stable-diffusion.cpp-rest/src/server.cpp

#include "server.h"
#include "model_manager.h"
#include "generation_queue.h"
#include "utils.h"
#include "auth_middleware.h"
#include "user_manager.h"
#include "version.h"
#include <httplib.h>
#include <nlohmann/json.hpp>
#include <iostream>
#include <sstream>
#include <fstream>
#include <chrono>
#include <random>
#include <algorithm>
#include <thread>
#include <filesystem>

// Include stb_image for loading images (implementation is in generation_queue.cpp)
#include "../stable-diffusion.cpp-src/thirdparty/stb_image.h"
#include <sys/socket.h>
#include <netinet/in.h>
#include <unistd.h>
#include <arpa/inet.h>


Server::Server(ModelManager* modelManager, GenerationQueue* generationQueue, const std::string& outputDir, const std::string& uiDir, const ServerConfig& config)
    : m_modelManager(modelManager)
    , m_generationQueue(generationQueue)
    , m_isRunning(false)
    , m_startupFailed(false)
    , m_port(config.port)
    , m_outputDir(outputDir)
    , m_uiDir(uiDir)
    , m_userManager(nullptr)
    , m_authMiddleware(nullptr)
    , m_config(config)
{
    m_httpServer = std::make_unique<httplib::Server>();
}

Server::~Server() {
    stop();
}

bool Server::start(const std::string& host, int port) {
    if (m_isRunning.load()) {
        return false;
    }

    m_host = host;
    m_port = port;


    // Validate host and port
    if (host.empty() || (port < 1 || port > 65535)) {
        return false;
    }

    // Set up CORS headers
    setupCORS();

    // Register API endpoints
    registerEndpoints();


    // Reset startup flags
    m_startupFailed.store(false);

    // Start server in a separate thread
    m_serverThread = std::thread(&Server::serverThreadFunction, this, host, port);

    // Wait for server to actually start and bind to the port

    // Give more time for server to actually start and bind
    for (int i = 0; i < 100; i++) { // Wait up to 10 seconds
        std::this_thread::sleep_for(std::chrono::milliseconds(100));

        // Check if startup failed early
        if (m_startupFailed.load()) {
            if (m_serverThread.joinable()) {
                m_serverThread.join();
            }
            return false;
        }

        if (m_isRunning.load()) {
            // Give it a moment more to ensure server is fully started
            std::this_thread::sleep_for(std::chrono::milliseconds(500));
            if (m_isRunning.load()) {
                return true;
            }
        }
    }

    if (m_isRunning.load()) {
        return true;
    } else {
        if (m_serverThread.joinable()) {
            m_serverThread.join();
        }
        return false;
    }
}

void Server::stop() {
    // Use atomic check to ensure thread safety
    bool wasRunning = m_isRunning.exchange(false);
    if (!wasRunning) {
        return; // Already stopped
    }

    if (m_httpServer) {
        m_httpServer->stop();

        // Give the server a moment to stop the blocking listen call
        std::this_thread::sleep_for(std::chrono::milliseconds(100));

        // If server thread is still running, try to force unblock the listen call
        // by making a quick connection to the server port
        if (m_serverThread.joinable()) {
            try {
                // Create a quick connection to interrupt the blocking listen
                httplib::Client client("127.0.0.1", m_port);
                client.set_connection_timeout(0, m_config.connectionTimeoutMs * 1000); // Convert ms to microseconds
                client.set_read_timeout(0, m_config.readTimeoutMs * 1000); // Convert ms to microseconds
                client.set_write_timeout(0, m_config.writeTimeoutMs * 1000); // Convert ms to microseconds
                auto res = client.Get("/api/health");
                // We don't care about the response, just trying to unblock
            } catch (...) {
                // Ignore any connection errors - we're just trying to unblock
            }
        }
    }

    if (m_serverThread.joinable()) {
        m_serverThread.join();
    }

}

bool Server::isRunning() const {
    return m_isRunning.load();
}

void Server::waitForStop() {
    if (m_serverThread.joinable()) {
        m_serverThread.join();
    }
}

void Server::registerEndpoints() {
    // Register authentication endpoints first (before applying middleware)
    registerAuthEndpoints();

    // Health check endpoint (public)
    m_httpServer->Get("/api/health", [this](const httplib::Request& req, httplib::Response& res) {
        handleHealthCheck(req, res);
    });

    // API status endpoint (public)
    m_httpServer->Get("/api/status", [this](const httplib::Request& req, httplib::Response& res) {
        handleApiStatus(req, res);
    });

    // Version information endpoint (public)
    m_httpServer->Get("/api/version", [this](const httplib::Request& req, httplib::Response& res) {
        handleVersion(req, res);
    });

    // Apply authentication middleware to protected endpoints
    auto withAuth = [this](std::function<void(const httplib::Request&, httplib::Response&)> handler) {
        return [this, handler](const httplib::Request& req, httplib::Response& res) {
            if (m_authMiddleware) {
                AuthContext authContext = m_authMiddleware->authenticate(req, res);
                if (!authContext.authenticated) {
                    m_authMiddleware->sendAuthError(res, authContext.errorMessage, authContext.errorCode);
                    return;
                }
            }
            handler(req, res);
        };
    };

    // Specialized generation endpoints (protected)
    m_httpServer->Post("/api/generate/text2img", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleText2Img(req, res);
    }));

    m_httpServer->Post("/api/generate/img2img", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleImg2Img(req, res);
    }));

    m_httpServer->Post("/api/generate/controlnet", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleControlNet(req, res);
    }));

    m_httpServer->Post("/api/generate/upscale", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleUpscale(req, res);
    }));

    m_httpServer->Post("/api/generate/inpainting", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleInpainting(req, res);
    }));

    // Utility endpoints (now protected - require authentication)
    m_httpServer->Get("/api/samplers", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleSamplers(req, res);
    }));

    m_httpServer->Get("/api/schedulers", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleSchedulers(req, res);
    }));

    m_httpServer->Get("/api/parameters", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleParameters(req, res);
    }));

    m_httpServer->Post("/api/validate", [this](const httplib::Request& req, httplib::Response& res) {
        handleValidate(req, res);
    });

    m_httpServer->Post("/api/estimate", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleEstimate(req, res);
    }));

    m_httpServer->Get("/api/config", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleConfig(req, res);
    }));

    m_httpServer->Get("/api/system", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleSystem(req, res);
    }));

    m_httpServer->Post("/api/system/restart", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleSystemRestart(req, res);
    }));

    // Models list endpoint (now protected - require authentication)
    m_httpServer->Get("/api/models", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleModelsList(req, res);
    }));

    // Model-specific endpoints
    m_httpServer->Get("/api/models/(.*)", [this](const httplib::Request& req, httplib::Response& res) {
        handleModelInfo(req, res);
    });

    m_httpServer->Post("/api/models/(.*)/load", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleLoadModelById(req, res);
    }));

    m_httpServer->Post("/api/models/(.*)/unload", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleUnloadModelById(req, res);
    }));

    // Model management endpoints (now protected - require authentication)
    m_httpServer->Get("/api/models/types", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleModelTypes(req, res);
    }));

    m_httpServer->Get("/api/models/directories", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleModelDirectories(req, res);
    }));

    m_httpServer->Post("/api/models/refresh", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleRefreshModels(req, res);
    }));

    m_httpServer->Post("/api/models/hash", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleHashModels(req, res);
    }));

    m_httpServer->Post("/api/models/convert", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleConvertModel(req, res);
    }));

    m_httpServer->Get("/api/models/stats", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleModelStats(req, res);
    }));

    m_httpServer->Post("/api/models/batch", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleBatchModels(req, res);
    }));

    // Model validation endpoints (already protected with withAuth)
    m_httpServer->Post("/api/models/validate", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleValidateModel(req, res);
    }));

    m_httpServer->Post("/api/models/compatible", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleCheckCompatibility(req, res);
    }));

    m_httpServer->Post("/api/models/requirements", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleModelRequirements(req, res);
    }));

    // Queue status endpoint (now protected - require authentication)
    m_httpServer->Get("/api/queue/status", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleQueueStatus(req, res);
    }));

    // Download job output file endpoint (must be before job status endpoint to match more specific pattern first)
    // Note: This endpoint is public to allow frontend to display generated images without authentication
    m_httpServer->Get("/api/queue/job/(.*)/output/(.*)", [this](const httplib::Request& req, httplib::Response& res) {
        handleDownloadOutput(req, res);
    });

    // Get job output by job ID endpoint (public to allow frontend to display generated images without authentication)
    m_httpServer->Get("/api/v1/jobs/(.*)/output", [this](const httplib::Request& req, httplib::Response& res) {
        handleJobOutput(req, res);
    });

    // Download image from URL endpoint (public for CORS-free image handling)
    m_httpServer->Get("/api/image/download", [this](const httplib::Request& req, httplib::Response& res) {
        handleDownloadImageFromUrl(req, res);
    });

    // Image resize endpoint (protected)
    m_httpServer->Post("/api/image/resize", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleImageResize(req, res);
    }));

    // Image crop endpoint (protected)
    m_httpServer->Post("/api/image/crop", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleImageCrop(req, res);
    }));

    // Job status endpoint (now protected - require authentication)
    m_httpServer->Get("/api/queue/job/(.*)", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleJobStatus(req, res);
    }));

    // Cancel job endpoint (protected)
    m_httpServer->Post("/api/queue/cancel", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleCancelJob(req, res);
    }));

    // Clear queue endpoint (protected)
    m_httpServer->Post("/api/queue/clear", withAuth([this](const httplib::Request& req, httplib::Response& res) {
        handleClearQueue(req, res);
    }));

    // Serve static web UI files if uiDir is configured
    if (!m_uiDir.empty() && std::filesystem::exists(m_uiDir)) {
        std::cout << "Serving static UI files from: " << m_uiDir << " at /ui" << std::endl;

        // Read UI version from version.nlohmann::json if available
        std::string uiVersion = "unknown";
        std::string versionFilePath = m_uiDir + "/version.nlohmann::json";
        if (std::filesystem::exists(versionFilePath)) {
            try {
                std::ifstream versionFile(versionFilePath);
                if (versionFile.is_open()) {
                    nlohmann::json versionData = nlohmann::json::parse(versionFile);
                    if (versionData.contains("version")) {
                        uiVersion = versionData["version"].get<std::string>();
                    }
                    versionFile.close();
                }
            } catch (const std::exception& e) {
                std::cerr << "Failed to read UI version: " << e.what() << std::endl;
            }
        }
        std::cout << "UI version: " << uiVersion << std::endl;

        // Serve dynamic config.js that provides runtime configuration to the web UI
        m_httpServer->Get("/ui/config.js", [this, uiVersion](const httplib::Request& /*req*/, httplib::Response& res) {
            // Generate JavaScript configuration with current server settings
            std::ostringstream configJs;
            configJs << "// Auto-generated configuration\n"
                     << "window.__SERVER_CONFIG__ = {\n"
                     << "  apiUrl: 'http://" << m_host << ":" << m_port << "',\n"
                     << "  apiBasePath: '/api',\n"
                     << "  host: '" << m_host << "',\n"
                     << "  port: " << m_port << ",\n"
                     << "  uiVersion: '" << uiVersion << "',\n";

            // Add authentication method information
            if (m_authMiddleware) {
                auto authConfig = m_authMiddleware->getConfig();
                std::string authMethod = "none";
                switch (authConfig.authMethod) {
                    case AuthMethod::UNIX:
                        authMethod = "unix";
                        break;
                    case AuthMethod::JWT:
                        authMethod = "jwt";
                        break;
                    default:
                        authMethod = "none";
                        break;
                }
                configJs << "  authMethod: '" << authMethod << "',\n"
                         << "  authEnabled: " << (authConfig.authMethod != AuthMethod::NONE ? "true" : "false") << "\n";
            } else {
                configJs << "  authMethod: 'none',\n"
                         << "  authEnabled: false\n";
            }

            configJs << "};\n";

            // No cache for config.js - always fetch fresh
            res.set_header("Cache-Control", "no-cache, no-store, must-revalidate");
            res.set_header("Pragma", "no-cache");
            res.set_header("Expires", "0");
            res.set_content(configJs.str(), "application/javascript");
        });

        // Set up file request handler for caching static assets
        m_httpServer->set_file_request_handler([uiVersion](const httplib::Request& req, httplib::Response& res) {
            // Add cache headers based on file type and version
            std::string path = req.path;

            // For versioned static assets (.js, .css, images), use long cache
            if (path.find("/_next/") != std::string::npos ||
                path.find(".js") != std::string::npos ||
                path.find(".css") != std::string::npos ||
                path.find(".png") != std::string::npos ||
                path.find(".jpg") != std::string::npos ||
                path.find(".svg") != std::string::npos ||
                path.find(".ico") != std::string::npos ||
                path.find(".woff") != std::string::npos ||
                path.find(".woff2") != std::string::npos ||
                path.find(".ttf") != std::string::npos) {

                // Long cache (1 year) for static assets
                res.set_header("Cache-Control", "public, max-age=31536000, immutable");

                // Add ETag based on UI version for cache validation
                res.set_header("ETag", "\"" + uiVersion + "\"");

                // Check If-None-Match for conditional requests
                if (req.has_header("If-None-Match")) {
                    std::string clientETag = req.get_header_value("If-None-Match");
                    if (clientETag == "\"" + uiVersion + "\"") {
                        res.status = 304; // Not Modified
                        return;
                    }
                }
            } else if (path.find(".html") != std::string::npos || path == "/ui/" || path == "/ui") {
                // HTML files should revalidate but can be cached briefly
                res.set_header("Cache-Control", "public, max-age=0, must-revalidate");
                res.set_header("ETag", "\"" + uiVersion + "\"");
            }
        });

        // Create a handler for UI routes with authentication check
        auto uiHandler = [this](const httplib::Request& req, httplib::Response& res) {
            // Check if authentication is enabled
            if (m_authMiddleware) {
                auto authConfig = m_authMiddleware->getConfig();
                if (authConfig.authMethod != AuthMethod::NONE) {
                    // Authentication is enabled, check if user is authenticated
                    AuthContext authContext = m_authMiddleware->authenticate(req, res);

                    // For Unix auth, we need to check if the user is authenticated
                    // The authenticateUnix function will return a guest context for UI requests
                    // when no Authorization header is present, but we still need to show the login page
                    if (!authContext.authenticated) {
                        // Check if this is a request for a static asset (JS, CSS, images)
                        // These should be served even without authentication to allow the login page to work
                        bool isStaticAsset = false;
                        std::string path = req.path;
                        if (path.find(".js") != std::string::npos ||
                            path.find(".css") != std::string::npos ||
                            path.find(".png") != std::string::npos ||
                            path.find(".jpg") != std::string::npos ||
                            path.find(".jpeg") != std::string::npos ||
                            path.find(".svg") != std::string::npos ||
                            path.find(".ico") != std::string::npos ||
                            path.find("/_next/") != std::string::npos) {
                            isStaticAsset = true;
                        }

                        // For static assets, allow them to be served without authentication
                        if (isStaticAsset) {
                            // Continue to serve the file
                        } else {
                            // For HTML requests, redirect to login page
                            if (req.path.find(".html") != std::string::npos ||
                                req.path == "/ui/" || req.path == "/ui") {
                                // Serve the login page instead of the requested page
                                std::string loginPagePath = m_uiDir + "/login.html";
                                if (std::filesystem::exists(loginPagePath)) {
                                    std::ifstream loginFile(loginPagePath);
                                    if (loginFile.is_open()) {
                                        std::string content((std::istreambuf_iterator<char>(loginFile)),
                                                             std::istreambuf_iterator<char>());
                                        res.set_content(content, "text/html");
                                        return;
                                    }
                                }
                                // If login.html doesn't exist, serve a simple login page
                                std::string simpleLoginPage = R"(
<!DOCTYPE html>
<html>
<head>
    <title>Login Required</title>
    <style>
        body { font-family: Arial, sans-serif; max-width: 500px; margin: 100px auto; padding: 20px; }
        .form-group { margin-bottom: 15px; }
        label { display: block; margin-bottom: 5px; }
        input { width: 100%; padding: 8px; box-sizing: border-box; }
        button { background-color: #007bff; color: white; padding: 10px 15px; border: none; cursor: pointer; }
        .error { color: red; margin-top: 10px; }
    </style>
</head>
<body>
    <h1>Login Required</h1>
    <p>Please enter your username to continue.</p>
    <form id="loginForm">
        <div class="form-group">
            <label for="username">Username:</label>
            <input type="text" id="username" name="username" required>
        </div>
        <button type="submit">Login</button>
    </form>
    <div id="error" class="error"></div>
    <script>
        document.getElementById('loginForm').addEventListener('submit', async (e) => {
            e.preventDefault();
            const username = document.getElementById('username').value;
            const errorDiv = document.getElementById('error');

            try {
                const response = await fetch('/api/auth/login', {
                    method: 'POST',
                    headers: { 'Content-Type': 'application/nlohmann::json' },
                    body: JSON.stringify({ username })
                });

                if (response.ok) {
                    const data = await response.nlohmann::json();
                    localStorage.setItem('auth_token', data.token);
                    localStorage.setItem('unix_user', username);
                    window.location.reload();
                } else {
                    const error = await response.nlohmann::json();
                    errorDiv.textContent = error.message || 'Login failed';
                }
            } catch (err) {
                errorDiv.textContent = 'Login failed: ' + err.message;
            }
        });
    </script>
</body>
</html>
)";
                                res.set_content(simpleLoginPage, "text/html");
                                return;
                            } else {
                                // For non-HTML files, return unauthorized
                                m_authMiddleware->sendAuthError(res, "Authentication required", "AUTH_REQUIRED");
                                return;
                            }
                        }
                    }
                }
            }

            // If we get here, either auth is disabled or user is authenticated
            // Serve the requested file
            std::string filePath = req.path.substr(3); // Remove "/ui" prefix
            if (filePath.empty() || filePath == "/") {
                filePath = "/index.html";
            }

            std::string fullPath = m_uiDir + filePath;
            if (std::filesystem::exists(fullPath) && std::filesystem::is_regular_file(fullPath)) {
                std::ifstream file(fullPath, std::ios::binary);
                if (file.is_open()) {
                    std::string content((std::istreambuf_iterator<char>(file)),
                                     std::istreambuf_iterator<char>());

                    // Determine content type based on file extension
                    std::string contentType = "text/plain";
                    if (filePath.find(".html") != std::string::npos) {
                        contentType = "text/html";
                    } else if (filePath.find(".js") != std::string::npos) {
                        contentType = "application/javascript";
                    } else if (filePath.find(".css") != std::string::npos) {
                        contentType = "text/css";
                    } else if (filePath.find(".png") != std::string::npos) {
                        contentType = "image/png";
                    } else if (filePath.find(".jpg") != std::string::npos || filePath.find(".jpeg") != std::string::npos) {
                        contentType = "image/jpeg";
                    } else if (filePath.find(".svg") != std::string::npos) {
                        contentType = "image/svg+xml";
                    }

                    res.set_content(content, contentType);
                } else {
                    res.status = 404;
                    res.set_content("File not found", "text/plain");
                }
            } else {
                // For SPA routing, if the file doesn't exist, serve index.html
                // This allows Next.js to handle client-side routing
                std::string indexPath = m_uiDir + "/index.html";
                if (std::filesystem::exists(indexPath)) {
                    std::ifstream indexFile(indexPath, std::ios::binary);
                    if (indexFile.is_open()) {
                        std::string content((std::istreambuf_iterator<char>(indexFile)),
                                         std::istreambuf_iterator<char>());
                        res.set_content(content, "text/html");
                    } else {
                        res.status = 404;
                        res.set_content("File not found", "text/plain");
                    }
                } else {
                    res.status = 404;
                    res.set_content("File not found", "text/plain");
                }
            }
        };

        // Set up UI routes with authentication
        m_httpServer->Get("/ui/.*", uiHandler);

        // Redirect /ui to /ui/ to ensure proper routing
        m_httpServer->Get("/ui", [](const httplib::Request& /*req*/, httplib::Response& res) {
            res.set_redirect("/ui/");
        });
    }
}

void Server::setAuthComponents(std::shared_ptr<UserManager> userManager, std::shared_ptr<AuthMiddleware> authMiddleware) {
    m_userManager = userManager;
    m_authMiddleware = authMiddleware;
}

void Server::registerAuthEndpoints() {
    // Login endpoint
    m_httpServer->Post("/api/auth/login", [this](const httplib::Request& req, httplib::Response& res) {
        handleLogin(req, res);
    });

    // Logout endpoint
    m_httpServer->Post("/api/auth/logout", [this](const httplib::Request& req, httplib::Response& res) {
        handleLogout(req, res);
    });

    // Token validation endpoint
    m_httpServer->Get("/api/auth/validate", [this](const httplib::Request& req, httplib::Response& res) {
        handleValidateToken(req, res);
    });

    // Refresh token endpoint
    m_httpServer->Post("/api/auth/refresh", [this](const httplib::Request& req, httplib::Response& res) {
        handleRefreshToken(req, res);
    });

    // Get current user endpoint
    m_httpServer->Get("/api/auth/me", [this](const httplib::Request& req, httplib::Response& res) {
        handleGetCurrentUser(req, res);
    });
}

void Server::handleLogin(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_userManager || !m_authMiddleware) {
            sendErrorResponse(res, "Authentication system not available", 500, "AUTH_UNAVAILABLE", requestId);
            return;
        }

        // Parse request body
        nlohmann::json requestJson;
        try {
            requestJson = nlohmann::json::parse(req.body);
        } catch (const nlohmann::json::parse_error& e) {
            sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
            return;
        }

        // Check if using Unix authentication
        if (m_authMiddleware->getConfig().authMethod == AuthMethod::UNIX) {
            // For Unix auth, get username and password from request body
            std::string username = requestJson.value("username", "");
            std::string password = requestJson.value("password", "");

            if (username.empty()) {
                sendErrorResponse(res, "Missing username", 400, "MISSING_USERNAME", requestId);
                return;
            }

            // Check if PAM is enabled - if so, password is required
            if (m_userManager->isPamAuthEnabled() && password.empty()) {
                sendErrorResponse(res, "Password is required for Unix authentication", 400, "MISSING_PASSWORD", requestId);
                return;
            }

            // Authenticate Unix user (with or without password depending on PAM)
            auto result = m_userManager->authenticateUnix(username, password);

            if (!result.success) {
                sendErrorResponse(res, result.errorMessage, 401, "UNIX_AUTH_FAILED", requestId);
                return;
            }

            // Generate simple token for Unix auth
            std::string token = "unix_token_" + std::to_string(std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()).count()) + "_" + username;

            nlohmann::json response = {
                {"token", token},
                {"user", {
                    {"id", result.userId},
                    {"username", result.username},
                    {"role", result.role},
                    {"permissions", result.permissions}
                }},
                {"message", "Unix authentication successful"}
            };

            sendJsonResponse(res, response);
            return;
        }

        // For non-Unix auth, validate required fields
        if (!requestJson.contains("username") || !requestJson.contains("password")) {
            sendErrorResponse(res, "Missing username or password", 400, "MISSING_CREDENTIALS", requestId);
            return;
        }

        std::string username = requestJson["username"];
        std::string password = requestJson["password"];

        // Authenticate user
        auto result = m_userManager->authenticateUser(username, password);

        if (!result.success) {
            sendErrorResponse(res, result.errorMessage, 401, "INVALID_CREDENTIALS", requestId);
            return;
        }

        // Generate JWT token if using JWT auth
        std::string token;
        if (m_authMiddleware->getConfig().authMethod == AuthMethod::JWT) {
            // For now, create a simple token (in a real implementation, use JWT)
            token = "token_" + std::to_string(std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()).count()) + "_" + username;
        }

        nlohmann::json response = {
            {"token", token},
            {"user", {
                {"id", result.userId},
                {"username", result.username},
                {"role", result.role},
                {"permissions", result.permissions}
            }},
            {"message", "Login successful"}
        };

        sendJsonResponse(res, response);

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Login failed: ") + e.what(), 500, "LOGIN_ERROR", requestId);
    }
}

void Server::handleLogout(const httplib::Request& /*req*/, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // For now, just return success (in a real implementation, invalidate the token)
        nlohmann::json response = {
            {"message", "Logout successful"}
        };

        sendJsonResponse(res, response);

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Logout failed: ") + e.what(), 500, "LOGOUT_ERROR", requestId);
    }
}

void Server::handleValidateToken(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_userManager || !m_authMiddleware) {
            sendErrorResponse(res, "Authentication system not available", 500, "AUTH_UNAVAILABLE", requestId);
            return;
        }

        // Extract token from header
        std::string authHeader = req.get_header_value("Authorization");
        if (authHeader.empty()) {
            sendErrorResponse(res, "Missing authorization token", 401, "MISSING_TOKEN", requestId);
            return;
        }

        // Simple token validation (in a real implementation, validate JWT)
        // For now, just check if it starts with "token_"
        if (authHeader.find("Bearer ") != 0) {
            sendErrorResponse(res, "Invalid authorization header format", 401, "INVALID_HEADER", requestId);
            return;
        }

        std::string token = authHeader.substr(7); // Remove "Bearer "

        if (token.find("token_") != 0) {
            sendErrorResponse(res, "Invalid token", 401, "INVALID_TOKEN", requestId);
            return;
        }

        // Extract username from token (simple format: token_timestamp_username)
        size_t last_underscore = token.find_last_of('_');
        if (last_underscore == std::string::npos) {
            sendErrorResponse(res, "Invalid token format", 401, "INVALID_TOKEN", requestId);
            return;
        }

        std::string username = token.substr(last_underscore + 1);

        // Get user info
        auto userInfo = m_userManager->getUserInfoByUsername(username);
        if (userInfo.id.empty()) {
            sendErrorResponse(res, "User not found", 401, "USER_NOT_FOUND", requestId);
            return;
        }

        nlohmann::json response = {
            {"user", {
                {"id", userInfo.id},
                {"username", userInfo.username},
                {"role", userInfo.role},
                {"permissions", userInfo.permissions}
            }},
            {"valid", true}
        };

        sendJsonResponse(res, response);

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Token validation failed: ") + e.what(), 500, "VALIDATION_ERROR", requestId);
    }
}

void Server::handleRefreshToken(const httplib::Request& /*req*/, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // For now, just return a new token (in a real implementation, refresh JWT)
        nlohmann::json response = {
            {"token", "new_token_" + std::to_string(std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()).count())},
            {"message", "Token refreshed successfully"}
        };

        sendJsonResponse(res, response);

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Token refresh failed: ") + e.what(), 500, "REFRESH_ERROR", requestId);
    }
}

void Server::handleGetCurrentUser(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_userManager || !m_authMiddleware) {
            sendErrorResponse(res, "Authentication system not available", 500, "AUTH_UNAVAILABLE", requestId);
            return;
        }

        // Authenticate the request
        AuthContext authContext = m_authMiddleware->authenticate(req, res);

        if (!authContext.authenticated) {
            sendErrorResponse(res, "Authentication required", 401, "AUTH_REQUIRED", requestId);
            return;
        }

        nlohmann::json response = {
            {"user", {
                {"id", authContext.userId},
                {"username", authContext.username},
                {"role", authContext.role},
                {"permissions", authContext.permissions}
            }}
        };

        sendJsonResponse(res, response);

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Get current user failed: ") + e.what(), 500, "USER_ERROR", requestId);
    }
}

void Server::setupCORS() {
    // Use post-routing handler to set CORS headers after the response is generated
    // This ensures we don't duplicate headers that may be set by other handlers
    m_httpServer->set_post_routing_handler([](const httplib::Request& /*req*/, httplib::Response& res) {
        // Only add CORS headers if they haven't been set already
        if (!res.has_header("Access-Control-Allow-Origin")) {
            res.set_header("Access-Control-Allow-Origin", "*");
        }
        if (!res.has_header("Access-Control-Allow-Methods")) {
            res.set_header("Access-Control-Allow-Methods", "GET, POST, PUT, DELETE, OPTIONS");
        }
        if (!res.has_header("Access-Control-Allow-Headers")) {
            res.set_header("Access-Control-Allow-Headers", "Content-Type, Authorization");
        }
    });

    // Handle OPTIONS requests for CORS preflight (API endpoints only)
    m_httpServer->Options("/api/.*", [](const httplib::Request&, httplib::Response& res) {
        res.set_header("Access-Control-Allow-Origin", "*");
        res.set_header("Access-Control-Allow-Methods", "GET, POST, PUT, DELETE, OPTIONS");
        res.set_header("Access-Control-Allow-Headers", "Content-Type, Authorization");
        res.status = 200;
    });
}

void Server::handleHealthCheck(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json response = {
            {"status", "healthy"},
            {"timestamp", std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()).count()},
            {"version", sd_rest::VERSION_INFO.version_full}
        };
        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Health check failed: ") + e.what(), 500);
    }
}

void Server::handleApiStatus(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json response = {
            {"server", {
                {"running", m_isRunning.load()},
                {"host", m_host},
                {"port", m_port}
            }},
            {"generation_queue", {
                {"running", m_generationQueue ? m_generationQueue->isRunning() : false},
                {"queue_size", m_generationQueue ? m_generationQueue->getQueueSize() : 0},
                {"active_generations", m_generationQueue ? m_generationQueue->getActiveGenerations() : 0}
            }},
            {"models", {
                {"loaded_count", m_modelManager ? m_modelManager->getLoadedModelsCount() : 0},
                {"available_count", m_modelManager ? m_modelManager->getAvailableModelsCount() : 0}
            }}
        };
        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Status check failed: ") + e.what(), 500);
    }
}

void Server::handleVersion(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json response = {
            {"version", sd_rest::VERSION_INFO.version_full},
            {"type", sd_rest::VERSION_INFO.version_type},
            {"commit", {
                {"short", sd_rest::VERSION_INFO.commit_short},
                {"full", sd_rest::VERSION_INFO.commit_full}
            }},
            {"branch", sd_rest::VERSION_INFO.branch},
            {"clean", sd_rest::VERSION_INFO.is_clean},
            {"build_time", sd_rest::VERSION_INFO.build_time}
        };
        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Version check failed: ") + e.what(), 500);
    }
}

// Helper function to convert ModelDetails vector to JSON array
nlohmann::json Server::modelDetailsToJson(const std::vector<ModelManager::ModelDetails>& modelDetails) {
    nlohmann::json jsonArray = nlohmann::json::array();
    
    for (const auto& detail : modelDetails) {
        nlohmann::json modelJson = {
            {"name", detail.name},
            {"exists", detail.exists},
            {"type", detail.type},
            {"file_size", detail.file_size}
        };
        
        // Handle path and sha256 separately to avoid type mismatch
        if (detail.exists) {
            modelJson["path"] = detail.path;
            modelJson["sha256"] = detail.sha256;
        } else {
            modelJson["path"] = nullptr;
            modelJson["sha256"] = "";
        }
        
        // Add conditional fields for required/recommended models
        if (detail.is_required) {
            modelJson["is_required"] = true;
        }
        if (detail.is_recommended) {
            modelJson["is_recommended"] = true;
        }
        
        jsonArray.push_back(modelJson);
    }
    
    return jsonArray;
}

// Helper function to determine which recommended fields to include based on architecture
std::map<std::string, bool> Server::getRecommendedModelFields(const std::string& architecture) {
    std::map<std::string, bool> recommendedFields;
    
    // Initialize all fields as false (will be set to null if not applicable)
    recommendedFields["recommended_vae"] = false;
    recommendedFields["recommended_clip_l"] = false;
    recommendedFields["recommended_clip_g"] = false;
    recommendedFields["recommended_t5xxl"] = false;
    recommendedFields["recommended_clip_vision"] = false;
    recommendedFields["recommended_qwen2vl"] = false;
    
    // Architecture-specific field inclusion based on actual architecture strings
    if (architecture.find("Stable Diffusion 1.5") != std::string::npos) {
        // SD 1.x: recommended_vae only
        recommendedFields["recommended_vae"] = true;
    } else if (architecture.find("Stable Diffusion XL") != std::string::npos) {
        // SDXL: recommended_vae only
        recommendedFields["recommended_vae"] = true;
    } else if (architecture.find("Modern Architecture") != std::string::npos ||
               architecture.find("Flux Dev") != std::string::npos ||
               architecture.find("Flux Chroma") != std::string::npos) {
        // FLUX/SD3/Modern Architecture: recommended_vae, recommended_clip_l, recommended_t5xxl
        recommendedFields["recommended_vae"] = true;
        recommendedFields["recommended_clip_l"] = true;
        recommendedFields["recommended_t5xxl"] = true;
    } else if (architecture.find("SD 3") != std::string::npos) {
        // SD3: recommended_vae, recommended_clip_l, recommended_clip_g, recommended_t5xxl
        recommendedFields["recommended_vae"] = true;
        recommendedFields["recommended_clip_l"] = true;
        recommendedFields["recommended_clip_g"] = true;
        recommendedFields["recommended_t5xxl"] = true;
    } else if (architecture.find("Wan") != std::string::npos) {
        // Wan models: recommended_vae, recommended_t5xxl, recommended_clip_vision
        recommendedFields["recommended_vae"] = true;
        recommendedFields["recommended_t5xxl"] = true;
        recommendedFields["recommended_clip_vision"] = true;
    } else if (architecture.find("Qwen") != std::string::npos) {
        // Qwen models: recommended_vae, recommended_qwen2vl
        recommendedFields["recommended_vae"] = true;
        recommendedFields["recommended_qwen2vl"] = true;
    }
    // For UNKNOWN architecture, keep all fields false
    
    return recommendedFields;
}

// Helper function to populate recommended models with existence information
void Server::populateRecommendedModels(nlohmann::json& response, const ModelManager::ModelInfo& modelInfo) {
    if (modelInfo.requiredModels.empty()) {
        return;
    }
    
    // Check existence of required models
    auto requiredModelsDetails = m_modelManager->checkRequiredModelsExistence(modelInfo.requiredModels);
    
    // Get the recommended fields for this architecture
    auto recommendedFields = getRecommendedModelFields(modelInfo.architecture);
    
    // Group models by type
    std::map<std::string, std::vector<ModelManager::ModelDetails>> modelsByType;
    for (const auto& detail : requiredModelsDetails) {
        modelsByType[detail.type].push_back(detail);
    }
    
    // Populate recommended fields based on model types and architecture requirements
    for (const auto& [type, models] : modelsByType) {
        if (type == "VAE" && recommendedFields["recommended_vae"]) {
            response["recommended_vae"] = modelDetailsToJson(models);
        } else if (type == "CLIP-L" && recommendedFields["recommended_clip_l"]) {
            response["recommended_clip_l"] = modelDetailsToJson(models);
        } else if (type == "CLIP-G" && recommendedFields["recommended_clip_g"]) {
            response["recommended_clip_g"] = modelDetailsToJson(models);
        } else if (type == "T5XXL" && recommendedFields["recommended_t5xxl"]) {
            response["recommended_t5xxl"] = modelDetailsToJson(models);
        } else if (type == "CLIP-Vision" && recommendedFields["recommended_clip_vision"]) {
            response["recommended_clip_vision"] = modelDetailsToJson(models);
        } else if (type == "Qwen2VL" && recommendedFields["recommended_qwen2vl"]) {
            response["recommended_qwen2vl"] = modelDetailsToJson(models);
        }
    }
    
    // Set non-applicable fields to null
    for (const auto& [fieldName, shouldInclude] : recommendedFields) {
        if (!shouldInclude || !response.contains(fieldName)) {
            response[fieldName] = nlohmann::json(nullptr);
        }
    }
}

void Server::handleModelsList(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Parse query parameters for enhanced filtering
        std::string typeFilter = req.get_param_value("type");
        std::string searchQuery = req.get_param_value("search");
        std::string sortBy = req.get_param_value("sort_by");
        std::string sortOrder = req.get_param_value("sort_order");
        std::string dateFilter = req.get_param_value("date");
        std::string sizeFilter = req.get_param_value("size");

        // Pagination parameters - only apply if limit is explicitly provided
        int page = 1;
        int limit = 50;
        bool usePagination = false;

        try {
            if (!req.get_param_value("limit").empty()) {
                limit = std::stoi(req.get_param_value("limit"));
                // Special case: limit<=0 means return all models (no pagination)
                if (limit <= 0) {
                    usePagination = false;
                    limit = INT_MAX; // Set to very large number to effectively disable pagination
                } else {
                    usePagination = true;
                    if (!req.get_param_value("page").empty()) {
                        page = std::stoi(req.get_param_value("page"));
                        if (page < 1) page = 1;
                    }
                }
            }
        } catch (const std::exception& e) {
            sendErrorResponse(res, "Invalid pagination parameters", 400, "INVALID_PAGINATION", requestId);
            return;
        }

        // Filter parameters
        bool includeLoaded = req.get_param_value("loaded") == "true";
        bool includeUnloaded = req.get_param_value("unloaded") == "true";
        (void)req.get_param_value("include_metadata"); // unused but kept for API compatibility
        (void)req.get_param_value("include_thumbnails"); // unused but kept for API compatibility

        // Get all models
        auto allModels = m_modelManager->getAllModels();
        nlohmann::json models = nlohmann::json::array();

        // Apply filters and build response
        for (const auto& pair : allModels) {
            const auto& modelInfo = pair.second;

            // Apply type filter
            if (!typeFilter.empty()) {
                ModelType filterType = ModelManager::stringToModelType(typeFilter);
                if (modelInfo.type != filterType) continue;
            }

            // Apply loaded/unloaded filters
            if (includeLoaded && !modelInfo.isLoaded) continue;
            if (includeUnloaded && modelInfo.isLoaded) continue;

            // Apply search filter (case-insensitive search in name and description)
            if (!searchQuery.empty()) {
                std::string searchLower = searchQuery;
                std::transform(searchLower.begin(), searchLower.end(), searchLower.begin(), ::tolower);

                std::string nameLower = modelInfo.name;
                std::transform(nameLower.begin(), nameLower.end(), nameLower.begin(), ::tolower);

                std::string descLower = modelInfo.description;
                std::transform(descLower.begin(), descLower.end(), descLower.begin(), ::tolower);

                if (nameLower.find(searchLower) == std::string::npos &&
                    descLower.find(searchLower) == std::string::npos) {
                    continue;
                }
            }

            // Apply date filter (simplified - expects "recent", "old", or YYYY-MM-DD)
            if (!dateFilter.empty()) {
                auto now = std::filesystem::file_time_type::clock::now();
                auto modelTime = modelInfo.modifiedAt;
                auto duration = std::chrono::duration_cast<std::chrono::hours>(now - modelTime).count();

                if (dateFilter == "recent" && duration > 24 * 7) continue; // Older than 1 week
                if (dateFilter == "old" && duration < 24 * 30) continue; // Newer than 1 month
            }

            // Apply size filter (expects "small", "medium", "large", or size in MB)
            if (!sizeFilter.empty()) {
                double sizeMB = modelInfo.fileSize / (1024.0 * 1024.0);

                if (sizeFilter == "small" && sizeMB > 1024) continue; // > 1GB
                if (sizeFilter == "medium" && (sizeMB < 1024 || sizeMB > 4096)) continue; // < 1GB or > 4GB
                if (sizeFilter == "large" && sizeMB < 4096) continue; // < 4GB

                // Try to parse as specific size in MB
                try {
                    double maxSizeMB = std::stod(sizeFilter);
                    if (sizeMB > maxSizeMB) continue;
                } catch (...) {
                    // Ignore if parsing fails
                }
            }

            // Build model JSON with enhanced structure
            nlohmann::json modelJson = {
                {"name", modelInfo.name},
                {"type", ModelManager::modelTypeToString(modelInfo.type)},
                {"file_size", modelInfo.fileSize},
                {"file_size_mb", modelInfo.fileSize / (1024.0 * 1024.0)},
                {"sha256", modelInfo.sha256.empty() ? nullptr : nlohmann::json(modelInfo.sha256)},
                {"sha256_short", (modelInfo.sha256.empty() || modelInfo.sha256.length() < 10) ? nullptr : nlohmann::json(modelInfo.sha256.substr(0, 10))}
            };

            // Add architecture information if available (checkpoints only)
            if (!modelInfo.architecture.empty()) {
                modelJson["architecture"] = modelInfo.architecture;
                modelJson["recommended_vae"] = modelInfo.recommendedVAE.empty() ? nullptr : nlohmann::json(modelInfo.recommendedVAE);

                if (modelInfo.recommendedWidth > 0) {
                    modelJson["recommended_width"] = modelInfo.recommendedWidth;
                }
                if (modelInfo.recommendedHeight > 0) {
                    modelJson["recommended_height"] = modelInfo.recommendedHeight;
                }
                if (modelInfo.recommendedSteps > 0) {
                    modelJson["recommended_steps"] = modelInfo.recommendedSteps;
                }
                if (!modelInfo.recommendedSampler.empty()) {
                    modelJson["recommended_sampler"] = modelInfo.recommendedSampler;
                }
                
                // Enhanced model information with existence checking
                if (!modelInfo.requiredModels.empty()) {
                    auto requiredModelsDetails = m_modelManager->checkRequiredModelsExistence(modelInfo.requiredModels);
                    modelJson["required_models"] = modelDetailsToJson(requiredModelsDetails);
                    
                    // Populate recommended models based on architecture
                    populateRecommendedModels(modelJson, modelInfo);
                }
                
                // Backward compatibility - keep existing fields
                if (!modelInfo.missingModels.empty()) {
                    modelJson["missing_models"] = modelInfo.missingModels;
                    modelJson["has_missing_dependencies"] = true;
                } else {
                    modelJson["has_missing_dependencies"] = false;
                }
            }

            models.push_back(modelJson);
        }

        // Apply sorting
        if (!sortBy.empty()) {
            std::sort(models.begin(), models.end(), [&sortBy, &sortOrder](const nlohmann::json& a, const nlohmann::json& b) {
                bool ascending = sortOrder != "desc";

                if (sortBy == "name") {
                    return ascending ? a["name"] < b["name"] : a["name"] > b["name"];
                } else if (sortBy == "size") {
                    return ascending ? a["file_size"] < b["file_size"] : a["file_size"] > b["file_size"];
                } else if (sortBy == "date") {
                    return ascending ? a["last_modified"] < b["last_modified"] : a["last_modified"] > b["last_modified"];
                } else if (sortBy == "type") {
                    return ascending ? a["type"] < b["type"] : a["type"] > b["type"];
                } else if (sortBy == "loaded") {
                    return ascending ? a["is_loaded"] < b["is_loaded"] : a["is_loaded"] > b["is_loaded"];
                }

                return false;
            });
        }

        // Apply pagination only if limit parameter was provided
        int totalCount = models.size();
        nlohmann::json paginatedModels = nlohmann::json::array();
        nlohmann::json paginationInfo = nlohmann::json::object();

        if (usePagination) {
            // Apply pagination
            int totalPages = (totalCount + limit - 1) / limit;
            int startIndex = (page - 1) * limit;
            int endIndex = std::min(startIndex + limit, totalCount);

            for (int i = startIndex; i < endIndex; ++i) {
                paginatedModels.push_back(models[i]);
            }

            paginationInfo = {
                {"page", page},
                {"limit", limit},
                {"total_count", totalCount},
                {"total_pages", totalPages},
                {"has_next", page < totalPages},
                {"has_prev", page > 1}
            };
        } else {
            // Return all models without pagination
            paginatedModels = models;

            paginationInfo = {
                {"page", 1},
                {"limit", totalCount},
                {"total_count", totalCount},
                {"total_pages", 1},
                {"has_next", false},
                {"has_prev", false}
            };
        }

        // Build comprehensive response
        nlohmann::json response = {
            {"models", paginatedModels},
            {"pagination", paginationInfo},
            {"filters_applied", {
                {"type", typeFilter.empty() ? nlohmann::json(nullptr) : nlohmann::json(typeFilter)},
                {"search", searchQuery.empty() ? nlohmann::json(nullptr) : nlohmann::json(searchQuery)},
                {"date", dateFilter.empty() ? nlohmann::json(nullptr) : nlohmann::json(dateFilter)},
                {"size", sizeFilter.empty() ? nlohmann::json(nullptr) : nlohmann::json(sizeFilter)},
                {"loaded", includeLoaded ? nlohmann::json(true) : nlohmann::json(nullptr)},
                {"unloaded", includeUnloaded ? nlohmann::json(true) : nlohmann::json(nullptr)}
            }},
            {"sorting", {
                {"sort_by", sortBy.empty() ? "name" : nlohmann::json(sortBy)},
                {"sort_order", sortOrder.empty() ? "asc" : nlohmann::json(sortOrder)}
            }},
            {"statistics", {
                {"loaded_count", m_modelManager->getLoadedModelsCount()},
                {"available_count", m_modelManager->getAvailableModelsCount()}
            }},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to list models: ") + e.what(), 500, "MODEL_LIST_ERROR", requestId);
    }
}


void Server::handleQueueStatus(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500);
            return;
        }

        // Get detailed queue status
        auto jobs = m_generationQueue->getQueueStatus();

        // Convert jobs to JSON
        nlohmann::json jobsJson = nlohmann::json::array();
        for (const auto& job : jobs) {
            std::string statusStr;
            switch (job.status) {
                case GenerationStatus::QUEUED: statusStr = "queued"; break;
                case GenerationStatus::PROCESSING: statusStr = "processing"; break;
                case GenerationStatus::COMPLETED: statusStr = "completed"; break;
                case GenerationStatus::FAILED: statusStr = "failed"; break;
            }

            // Convert time points to timestamps
            auto queuedTime = std::chrono::duration_cast<std::chrono::milliseconds>(
                job.queuedTime.time_since_epoch()).count();
            auto startTime = std::chrono::duration_cast<std::chrono::milliseconds>(
                job.startTime.time_since_epoch()).count();
            auto endTime = std::chrono::duration_cast<std::chrono::milliseconds>(
                job.endTime.time_since_epoch()).count();

            jobsJson.push_back({
                {"id", job.id},
                {"status", statusStr},
                {"prompt", job.prompt},
                {"queued_time", queuedTime},
                {"start_time", startTime > 0 ? nlohmann::json(startTime) : nlohmann::json(nullptr)},
                {"end_time", endTime > 0 ? nlohmann::json(endTime) : nlohmann::json(nullptr)},
                {"position", job.position},
                {"progress", job.progress}
            });
        }

        nlohmann::json response = {
            {"queue", {
                {"size", m_generationQueue->getQueueSize()},
                {"active_generations", m_generationQueue->getActiveGenerations()},
                {"running", m_generationQueue->isRunning()},
                {"jobs", jobsJson}
            }}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Queue status check failed: ") + e.what(), 500);
    }
}

void Server::handleJobStatus(const httplib::Request& req, httplib::Response& res) {
    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500);
            return;
        }

        // Extract job ID from URL path
        std::string jobId = req.matches[1].str();
        if (jobId.empty()) {
            sendErrorResponse(res, "Missing job ID", 400);
            return;
        }

        // Get job information
        auto jobInfo = m_generationQueue->getJobInfo(jobId);

        if (jobInfo.id.empty()) {
            sendErrorResponse(res, "Job not found", 404);
            return;
        }

        // Convert status to string
        std::string statusStr;
        switch (jobInfo.status) {
            case GenerationStatus::QUEUED: statusStr = "queued"; break;
            case GenerationStatus::PROCESSING: statusStr = "processing"; break;
            case GenerationStatus::COMPLETED: statusStr = "completed"; break;
            case GenerationStatus::FAILED: statusStr = "failed"; break;
        }

        // Convert time points to timestamps
        auto queuedTime = std::chrono::duration_cast<std::chrono::milliseconds>(
            jobInfo.queuedTime.time_since_epoch()).count();
        auto startTime = std::chrono::duration_cast<std::chrono::milliseconds>(
            jobInfo.startTime.time_since_epoch()).count();
        auto endTime = std::chrono::duration_cast<std::chrono::milliseconds>(
            jobInfo.endTime.time_since_epoch()).count();

        // Create download URLs for output files
        nlohmann::json outputUrls = nlohmann::json::array();
        for (const auto& filePath : jobInfo.outputFiles) {
            // Extract filename from full path
            std::filesystem::path p(filePath);
            std::string filename = p.filename().string();

            // Create download URL
            std::string url = "/api/queue/job/" + jobInfo.id + "/output/" + filename;

            nlohmann::json fileInfo = {
                {"filename", filename},
                {"url", url},
                {"path", filePath}
            };
            outputUrls.push_back(fileInfo);
        }

        nlohmann::json response = {
            {"job", {
                {"id", jobInfo.id},
                {"status", statusStr},
                {"prompt", jobInfo.prompt},
                {"queued_time", queuedTime},
                {"start_time", startTime > 0 ? nlohmann::json(startTime) : nlohmann::json(nullptr)},
                {"end_time", endTime > 0 ? nlohmann::json(endTime) : nlohmann::json(nullptr)},
                {"position", jobInfo.position},
                {"outputs", outputUrls},
                {"error_message", jobInfo.errorMessage},
                {"progress", jobInfo.progress}
            }}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Job status check failed: ") + e.what(), 500);
    }
}

void Server::handleCancelJob(const httplib::Request& req, httplib::Response& res) {
    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500);
            return;
        }

        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields
        if (!requestJson.contains("job_id") || !requestJson["job_id"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'job_id' field", 400);
            return;
        }

        std::string jobId = requestJson["job_id"];

        // Try to cancel the job
        bool cancelled = m_generationQueue->cancelJob(jobId);

        if (cancelled) {
            nlohmann::json response = {
                {"status", "success"},
                {"message", "Job cancelled successfully"},
                {"job_id", jobId}
            };
            sendJsonResponse(res, response);
        } else {
            nlohmann::json response = {
                {"status", "error"},
                {"message", "Job not found or already processing"},
                {"job_id", jobId}
            };
            sendJsonResponse(res, response, 404);
        }
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Job cancellation failed: ") + e.what(), 500);
    }
}

void Server::handleClearQueue(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500);
            return;
        }

        // Clear the queue
        m_generationQueue->clearQueue();

        nlohmann::json response = {
            {"status", "success"},
            {"message", "Queue cleared successfully"}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Queue clear failed: ") + e.what(), 500);
    }
}


void Server::handleDownloadOutput(const httplib::Request& req, httplib::Response& res) {
    try {
        // Extract job ID and filename from URL path
        if (req.matches.size() < 3) {
            sendErrorResponse(res, "Invalid request: job ID and filename required", 400, "INVALID_REQUEST", "");
            return;
        }

        std::string jobId = req.matches[1];
        std::string filename = req.matches[2];

        // Validate inputs
        if (jobId.empty() || filename.empty()) {
            sendErrorResponse(res, "Job ID and filename cannot be empty", 400, "INVALID_PARAMETERS", "");
            return;
        }

        // Construct absolute file path using the same logic as when saving:
        // {outputDir}/{jobId}/{filename}
        std::string fullPath = std::filesystem::absolute(m_outputDir + "/" + jobId + "/" + filename).string();

        // Log the request for debugging
        std::cout << "Image download request: jobId=" << jobId << ", filename=" << filename
                  << ", fullPath=" << fullPath << std::endl;

        // Check if file exists
        if (!std::filesystem::exists(fullPath)) {
            std::cerr << "Output file not found: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file not found: " + filename, 404, "FILE_NOT_FOUND", "");
            return;
        }

        // Check file size to detect zero-byte files
        auto fileSize = std::filesystem::file_size(fullPath);
        if (fileSize == 0) {
            std::cerr << "Output file is zero bytes: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file is empty (corrupted generation)", 500, "EMPTY_FILE", "");
            return;
        }

        // Check if file is accessible
        std::ifstream file(fullPath, std::ios::binary);
        if (!file.is_open()) {
            std::cerr << "Failed to open output file: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file not accessible", 500, "FILE_ACCESS_ERROR", "");
            return;
        }

        // Read file contents
        std::string fileContent;
        try {
            fileContent = std::string(
                std::istreambuf_iterator<char>(file),
                std::istreambuf_iterator<char>()
            );
            file.close();
        } catch (const std::exception& e) {
            std::cerr << "Failed to read file content: " << e.what() << std::endl;
            sendErrorResponse(res, "Failed to read file content", 500, "FILE_READ_ERROR", "");
            return;
        }

        // Verify we actually read data
        if (fileContent.empty()) {
            std::cerr << "File content is empty after read: " << fullPath << std::endl;
            sendErrorResponse(res, "File content is empty after read", 500, "EMPTY_CONTENT", "");
            return;
        }

        // Determine content type based on file extension
        std::string contentType = "application/octet-stream";

        if (Utils::endsWith(filename, ".png")) {
            contentType = "image/png";
        } else if (Utils::endsWith(filename, ".jpg") || Utils::endsWith(filename, ".jpeg")) {
            contentType = "image/jpeg";
        } else if (Utils::endsWith(filename, ".mp4")) {
            contentType = "video/mp4";
        } else if (Utils::endsWith(filename, ".gif")) {
            contentType = "image/gif";
        } else if (Utils::endsWith(filename, ".webp")) {
            contentType = "image/webp";
        }

        // Set response headers for proper browser handling
        res.set_header("Content-Type", contentType);
        res.set_header("Content-Length", std::to_string(fileContent.length()));
        res.set_header("Cache-Control", "public, max-age=3600"); // Cache for 1 hour
        res.set_header("Access-Control-Allow-Origin", "*"); // CORS for image access
        // Uncomment if you want to force download instead of inline display:
        // res.set_header("Content-Disposition", "attachment; filename=\"" + filename + "\"");

        // Set the content
        res.set_content(fileContent, contentType);
        res.status = 200;

        std::cout << "Successfully served image: " << filename << " (" << fileContent.length() << " bytes)" << std::endl;

    } catch (const std::exception& e) {
        std::cerr << "Exception in handleDownloadOutput: " << e.what() << std::endl;
        sendErrorResponse(res, std::string("Failed to download file: ") + e.what(), 500, "DOWNLOAD_ERROR", "");
    }
}

void Server::handleJobOutput(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();
    
    try {
        // Extract job ID from URL path
        if (req.matches.size() < 2) {
            sendErrorResponse(res, "Invalid request: job ID required", 400, "INVALID_REQUEST", requestId);
            return;
        }

        std::string jobId = req.matches[1].str();

        // Validate job ID
        if (jobId.empty()) {
            sendErrorResponse(res, "Job ID cannot be empty", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Log the request for debugging
        std::cout << "Job output request: jobId=" << jobId << std::endl;

        // Get job information to check if it exists and is completed
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        auto jobInfo = m_generationQueue->getJobInfo(jobId);
        if (jobInfo.id.empty()) {
            sendErrorResponse(res, "Job not found", 404, "JOB_NOT_FOUND", requestId);
            return;
        }

        // Check if job is completed
        if (jobInfo.status != GenerationStatus::COMPLETED) {
            std::string statusStr;
            switch (jobInfo.status) {
                case GenerationStatus::QUEUED: statusStr = "queued"; break;
                case GenerationStatus::PROCESSING: statusStr = "processing"; break;
                case GenerationStatus::FAILED: statusStr = "failed"; break;
                default: statusStr = "unknown"; break;
            }
            
            nlohmann::json response = {
                {"error", {
                    {"message", "Job not completed yet"},
                    {"status_code", 400},
                    {"error_code", "JOB_NOT_COMPLETED"},
                    {"request_id", requestId},
                    {"timestamp", std::chrono::duration_cast<std::chrono::seconds>(
                        std::chrono::system_clock::now().time_since_epoch()).count()},
                    {"job_status", statusStr}
                }}
            };
            sendJsonResponse(res, response, 400);
            return;
        }

        // Check if job has output files
        if (jobInfo.outputFiles.empty()) {
            sendErrorResponse(res, "No output files found for completed job", 404, "NO_OUTPUT_FILES", requestId);
            return;
        }

        // For simplicity, return the first output file
        // In a more complex implementation, we could return all files or allow file selection
        std::string firstOutputFile = jobInfo.outputFiles[0];
        
        // Extract filename from full path
        std::filesystem::path filePath(firstOutputFile);
        std::string filename = filePath.filename().string();
        
        // Construct absolute file path
        std::string fullPath = std::filesystem::absolute(firstOutputFile).string();

        // Check if file exists
        if (!std::filesystem::exists(fullPath)) {
            std::cerr << "Output file not found: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file not found: " + filename, 404, "FILE_NOT_FOUND", requestId);
            return;
        }

        // Check file size to detect zero-byte files
        auto fileSize = std::filesystem::file_size(fullPath);
        if (fileSize == 0) {
            std::cerr << "Output file is zero bytes: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file is empty (corrupted generation)", 500, "EMPTY_FILE", requestId);
            return;
        }

        // Check if file is accessible
        std::ifstream file(fullPath, std::ios::binary);
        if (!file.is_open()) {
            std::cerr << "Failed to open output file: " << fullPath << std::endl;
            sendErrorResponse(res, "Output file not accessible", 500, "FILE_ACCESS_ERROR", requestId);
            return;
        }

        // Read file contents
        std::string fileContent;
        try {
            fileContent = std::string(
                std::istreambuf_iterator<char>(file),
                std::istreambuf_iterator<char>()
            );
            file.close();
        } catch (const std::exception& e) {
            std::cerr << "Failed to read file content: " << e.what() << std::endl;
            sendErrorResponse(res, "Failed to read file content", 500, "FILE_READ_ERROR", requestId);
            return;
        }

        // Verify we actually read data
        if (fileContent.empty()) {
            std::cerr << "File content is empty after read: " << fullPath << std::endl;
            sendErrorResponse(res, "File content is empty after read", 500, "EMPTY_CONTENT", requestId);
            return;
        }

        // Determine content type based on file extension
        std::string contentType = "application/octet-stream";

        if (Utils::endsWith(filename, ".png")) {
            contentType = "image/png";
        } else if (Utils::endsWith(filename, ".jpg") || Utils::endsWith(filename, ".jpeg")) {
            contentType = "image/jpeg";
        } else if (Utils::endsWith(filename, ".mp4")) {
            contentType = "video/mp4";
        } else if (Utils::endsWith(filename, ".gif")) {
            contentType = "image/gif";
        } else if (Utils::endsWith(filename, ".webp")) {
            contentType = "image/webp";
        }

        // Set response headers for proper browser handling
        res.set_header("Content-Type", contentType);
        res.set_header("Content-Length", std::to_string(fileContent.length()));
        res.set_header("Cache-Control", "public, max-age=3600"); // Cache for 1 hour
        res.set_header("Access-Control-Allow-Origin", "*"); // CORS for image access
        
        // Set additional metadata headers
        res.set_header("X-Job-ID", jobId);
        res.set_header("X-Filename", filename);
        res.set_header("X-File-Size", std::to_string(fileSize));
        
        // If there are multiple files, indicate this
        if (jobInfo.outputFiles.size() > 1) {
            res.set_header("X-Total-Files", std::to_string(jobInfo.outputFiles.size()));
            res.set_header("X-File-Index", "1");
        }

        // Set the content
        res.set_content(fileContent, contentType);
        res.status = 200;

        std::cout << "Successfully served job output: jobId=" << jobId
                  << ", filename=" << filename
                  << " (" << fileContent.length() << " bytes)" << std::endl;

    } catch (const std::exception& e) {
        std::cerr << "Exception in handleJobOutput: " << e.what() << std::endl;
        sendErrorResponse(res, std::string("Failed to get job output: ") + e.what(), 500, "OUTPUT_ERROR", requestId);
    }
}

void Server::handleImageResize(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields
        if (!requestJson.contains("image") || !requestJson["image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("width") || !requestJson["width"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'width' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("height") || !requestJson["height"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'height' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        std::string imageInput = requestJson["image"];
        int targetWidth = requestJson["width"];
        int targetHeight = requestJson["height"];

        // Validate dimensions
        if (targetWidth < 1 || targetWidth > 4096) {
            sendErrorResponse(res, "Width must be between 1 and 4096", 400, "INVALID_DIMENSIONS", requestId);
            return;
        }

        if (targetHeight < 1 || targetHeight > 4096) {
            sendErrorResponse(res, "Height must be between 1 and 4096", 400, "INVALID_DIMENSIONS", requestId);
            return;
        }

        // Load the source image
        auto [imageData, sourceWidth, sourceHeight, sourceChannels, success, loadError] = loadImageFromInput(imageInput);

        if (!success) {
            sendErrorResponse(res, "Failed to load image: " + loadError, 400, "IMAGE_LOAD_ERROR", requestId);
            return;
        }

        // Convert image data to stb_image format for processing
        int channels = 3; // Force RGB
        size_t sourceSize = sourceWidth * sourceHeight * channels;
        std::vector<uint8_t> sourcePixels(sourceSize);
        std::memcpy(sourcePixels.data(), imageData.data(), std::min(imageData.size(), sourceSize));

        // Resize the image using stb_image_resize if available, otherwise use simple scaling
        std::vector<uint8_t> resizedPixels(targetWidth * targetHeight * channels);

        // Simple nearest-neighbor scaling for now (can be improved with better algorithms)
        float xScale = static_cast<float>(sourceWidth) / targetWidth;
        float yScale = static_cast<float>(sourceHeight) / targetHeight;

        for (int y = 0; y < targetHeight; y++) {
            for (int x = 0; x < targetWidth; x++) {
                int sourceX = static_cast<int>(x * xScale);
                int sourceY = static_cast<int>(y * yScale);

                // Clamp to source bounds
                sourceX = std::min(sourceX, sourceWidth - 1);
                sourceY = std::min(sourceY, sourceHeight - 1);

                for (int c = 0; c < channels; c++) {
                    resizedPixels[(y * targetWidth + x) * channels + c] =
                        sourcePixels[(sourceY * sourceWidth + sourceX) * channels + c];
                }
            }
        }

        // Convert resized image to base64
        std::string base64Data = Utils::base64Encode(resizedPixels);

        // Determine MIME type based on input
        std::string mimeType = "image/jpeg"; // default
        if (Utils::startsWith(imageInput, "data:image/png")) {
            mimeType = "image/png";
        } else if (Utils::startsWith(imageInput, "data:image/gif")) {
            mimeType = "image/gif";
        } else if (Utils::startsWith(imageInput, "data:image/webp")) {
            mimeType = "image/webp";
        } else if (Utils::startsWith(imageInput, "data:image/bmp")) {
            mimeType = "image/bmp";
        }

        // Create data URL format
        std::string dataUrl = "data:" + mimeType + ";base64," + base64Data;

        // Build response
        nlohmann::json response = {
            {"success", true},
            {"original_width", sourceWidth},
            {"original_height", sourceHeight},
            {"resized_width", targetWidth},
            {"resized_height", targetHeight},
            {"mime_type", mimeType},
            {"base64_data", dataUrl},
            {"file_size_bytes", resizedPixels.size()},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response, 200);

        std::cout << "Successfully resized image from " << sourceWidth << "x" << sourceHeight
                  << " to " << targetWidth << "x" << targetHeight
                  << " (" << resizedPixels.size() << " bytes)" << std::endl;

    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        std::cerr << "Exception in handleImageResize: " << e.what() << std::endl;
        sendErrorResponse(res, std::string("Failed to resize image: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleImageCrop(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields
        if (!requestJson.contains("image") || !requestJson["image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("x") || !requestJson["x"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'x' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("y") || !requestJson["y"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'y' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("width") || !requestJson["width"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'width' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        if (!requestJson.contains("height") || !requestJson["height"].is_number_integer()) {
            sendErrorResponse(res, "Missing or invalid 'height' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        std::string imageInput = requestJson["image"];
        int cropX = requestJson["x"];
        int cropY = requestJson["y"];
        int cropWidth = requestJson["width"];
        int cropHeight = requestJson["height"];

        // Load the source image
        auto [imageData, sourceWidth, sourceHeight, sourceChannels, success, loadError] = loadImageFromInput(imageInput);

        if (!success) {
            sendErrorResponse(res, "Failed to load image: " + loadError, 400, "IMAGE_LOAD_ERROR", requestId);
            return;
        }

        // Validate crop dimensions
        if (cropX < 0 || cropY < 0) {
            sendErrorResponse(res, "Crop coordinates must be non-negative", 400, "INVALID_CROP_AREA", requestId);
            return;
        }

        if (cropX + cropWidth > sourceWidth || cropY + cropHeight > sourceHeight) {
            sendErrorResponse(res, "Crop area exceeds image dimensions", 400, "INVALID_CROP_AREA", requestId);
            return;
        }

        if (cropWidth < 1 || cropHeight < 1) {
            sendErrorResponse(res, "Crop width and height must be at least 1", 400, "INVALID_CROP_AREA", requestId);
            return;
        }

        // Convert image data to stb_image format for processing
        int channels = 3; // Force RGB
        size_t sourceSize = sourceWidth * sourceHeight * channels;
        std::vector<uint8_t> sourcePixels(sourceSize);
        std::memcpy(sourcePixels.data(), imageData.data(), std::min(imageData.size(), sourceSize));

        // Crop the image
        std::vector<uint8_t> croppedPixels(cropWidth * cropHeight * channels);

        for (int y = 0; y < cropHeight; y++) {
            for (int x = 0; x < cropWidth; x++) {
                int sourceX = cropX + x;
                int sourceY = cropY + y;

                for (int c = 0; c < channels; c++) {
                    croppedPixels[(y * cropWidth + x) * channels + c] =
                        sourcePixels[(sourceY * sourceWidth + sourceX) * channels + c];
                }
            }
        }

        // Convert cropped image to base64
        std::string base64Data = Utils::base64Encode(croppedPixels);

        // Determine MIME type based on input
        std::string mimeType = "image/jpeg"; // default
        if (Utils::startsWith(imageInput, "data:image/png")) {
            mimeType = "image/png";
        } else if (Utils::startsWith(imageInput, "data:image/gif")) {
            mimeType = "image/gif";
        } else if (Utils::startsWith(imageInput, "data:image/webp")) {
            mimeType = "image/webp";
        } else if (Utils::startsWith(imageInput, "data:image/bmp")) {
            mimeType = "image/bmp";
        }

        // Create data URL format
        std::string dataUrl = "data:" + mimeType + ";base64," + base64Data;

        // Build response
        nlohmann::json response = {
            {"success", true},
            {"original_width", sourceWidth},
            {"original_height", sourceHeight},
            {"crop_x", cropX},
            {"crop_y", cropY},
            {"cropped_width", cropWidth},
            {"cropped_height", cropHeight},
            {"mime_type", mimeType},
            {"base64_data", dataUrl},
            {"file_size_bytes", croppedPixels.size()},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response, 200);

        std::cout << "Successfully cropped image from " << sourceWidth << "x" << sourceHeight
                  << " to " << cropWidth << "x" << cropHeight
                  << " at (" << cropX << "," << cropY << ")"
                  << " (" << croppedPixels.size() << " bytes)" << std::endl;

    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        std::cerr << "Exception in handleImageCrop: " << e.what() << std::endl;
        sendErrorResponse(res, std::string("Failed to crop image: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleDownloadImageFromUrl(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Parse query parameters
        std::string imageUrl = req.get_param_value("url");

        if (imageUrl.empty()) {
            sendErrorResponse(res, "Missing 'url' parameter", 400, "MISSING_URL", requestId);
            return;
        }

        // Basic URL format validation
        if (!Utils::startsWith(imageUrl, "http://") && !Utils::startsWith(imageUrl, "https://")) {
            sendErrorResponse(res, "Invalid URL format. URL must start with http:// or https://", 400, "INVALID_URL_FORMAT", requestId);
            return;
        }

        // Extract filename from URL for content type detection
        std::string filename = imageUrl;
        size_t lastSlash = imageUrl.find_last_of('/');
        if (lastSlash != std::string::npos) {
            filename = imageUrl.substr(lastSlash + 1);
        }

        // Remove query parameters and fragments
        size_t questionMark = filename.find('?');
        if (questionMark != std::string::npos) {
            filename = filename.substr(0, questionMark);
        }
        size_t hashMark = filename.find('#');
        if (hashMark != std::string::npos) {
            filename = filename.substr(0, hashMark);
        }

        // Check if URL has image extension
        std::string extension;
        size_t lastDot = filename.find_last_of('.');
        if (lastDot != std::string::npos) {
            extension = filename.substr(lastDot + 1);
            std::transform(extension.begin(), extension.end(), extension.begin(), ::tolower);
        }

        // Validate image extension
        const std::vector<std::string> validExtensions = {"jpg", "jpeg", "png", "gif", "webp", "bmp"};
        if (extension.empty() || std::find(validExtensions.begin(), validExtensions.end(), extension) == validExtensions.end()) {
            sendErrorResponse(res, "URL must point to an image file with a valid extension: " +
                             std::accumulate(validExtensions.begin(), validExtensions.end(), std::string(),
                                           [](const std::string& a, const std::string& b) {
                                               return a.empty() ? b : a + ", " + b;
                                           }), 400, "INVALID_IMAGE_EXTENSION", requestId);
            return;
        }

        // Load image using existing loadImageFromInput function
        auto [imageData, width, height, channels, success, error] = loadImageFromInput(imageUrl);

        if (!success) {
            sendErrorResponse(res, "Failed to download image from URL: " + error, 400, "IMAGE_DOWNLOAD_FAILED", requestId);
            return;
        }

        // Convert image data to base64
        std::string base64Data = Utils::base64Encode(imageData);

        // Determine MIME type based on extension
        std::string mimeType = "image/jpeg"; // default
        if (extension == "png") {
            mimeType = "image/png";
        } else if (extension == "gif") {
            mimeType = "image/gif";
        } else if (extension == "webp") {
            mimeType = "image/webp";
        } else if (extension == "bmp") {
            mimeType = "image/bmp";
        } else if (extension == "jpg" || extension == "jpeg") {
            mimeType = "image/jpeg";
        }

        // Create data URL format
        std::string dataUrl = "data:" + mimeType + ";base64," + base64Data;

        // Build response
        nlohmann::json response = {
            {"success", true},
            {"url", imageUrl},
            {"filename", filename},
            {"width", width},
            {"height", height},
            {"channels", channels},
            {"mime_type", mimeType},
            {"base64_data", dataUrl},
            {"file_size_bytes", imageData.size()},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response, 200);

        std::cout << "Successfully downloaded and encoded image from URL: " << imageUrl
                  << " (" << width << "x" << height << ", " << imageData.size() << " bytes)" << std::endl;

    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        std::cerr << "Exception in handleDownloadImageFromUrl: " << e.what() << std::endl;
        sendErrorResponse(res, std::string("Failed to download image from URL: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::sendJsonResponse(httplib::Response& res, const nlohmann::json& json, int status_code) {
    res.set_header("Content-Type", "application/json");
    res.status = status_code;
    res.body = json.dump();
}

void Server::sendErrorResponse(httplib::Response& res, const std::string& message, int status_code,
                              const std::string& error_code, const std::string& request_id) {
    nlohmann::json errorResponse = {
        {"error", {
            {"message", message},
            {"status_code", status_code},
            {"error_code", error_code},
            {"request_id", request_id},
            {"timestamp", std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()).count()}
        }}
    };
    sendJsonResponse(res, errorResponse, status_code);
}

std::pair<bool, std::string> Server::validateGenerationParameters(const nlohmann::json& params) {
    // Validate required fields
    if (!params.contains("prompt") || !params["prompt"].is_string()) {
        return {false, "Missing or invalid 'prompt' field"};
    }

    const std::string& prompt = params["prompt"];
    if (prompt.empty()) {
        return {false, "Prompt cannot be empty"};
    }
    if (prompt.length() > m_config.maxPromptLength) {
        return {false, "Prompt too long (max " + std::to_string(m_config.maxPromptLength) + " characters)"};
    }

    // Validate negative prompt if present
    if (params.contains("negative_prompt")) {
        if (!params["negative_prompt"].is_string()) {
            return {false, "Invalid 'negative_prompt' field, must be string"};
        }
        if (params["negative_prompt"].get<std::string>().length() > m_config.maxNegativePromptLength) {
            return {false, "Negative prompt too long (max " + std::to_string(m_config.maxNegativePromptLength) + " characters)"};
        }
    }

    // Validate width
    if (params.contains("width")) {
        if (!params["width"].is_number_integer()) {
            return {false, "Invalid 'width' field, must be integer"};
        }
        int width = params["width"];
        if (width < 64 || width > 2048 || width % 64 != 0) {
            return {false, "Width must be between 64 and 2048 and divisible by 64"};
        }
    }

    // Validate height
    if (params.contains("height")) {
        if (!params["height"].is_number_integer()) {
            return {false, "Invalid 'height' field, must be integer"};
        }
        int height = params["height"];
        if (height < 64 || height > 2048 || height % 64 != 0) {
            return {false, "Height must be between 64 and 2048 and divisible by 64"};
        }
    }

    // Validate batch count
    if (params.contains("batch_count")) {
        if (!params["batch_count"].is_number_integer()) {
            return {false, "Invalid 'batch_count' field, must be integer"};
        }
        int batchCount = params["batch_count"];
        if (batchCount < 1 || batchCount > 100) {
            return {false, "Batch count must be between 1 and 100"};
        }
    }

    // Validate steps
    if (params.contains("steps")) {
        if (!params["steps"].is_number_integer()) {
            return {false, "Invalid 'steps' field, must be integer"};
        }
        int steps = params["steps"];
        if (steps < 1 || steps > 150) {
            return {false, "Steps must be between 1 and 150"};
        }
    }

    // Validate CFG scale
    if (params.contains("cfg_scale")) {
        if (!params["cfg_scale"].is_number()) {
            return {false, "Invalid 'cfg_scale' field, must be number"};
        }
        float cfgScale = params["cfg_scale"];
        if (cfgScale < 1.0f || cfgScale > 30.0f) {
            return {false, "CFG scale must be between 1.0 and 30.0"};
        }
    }

    // Validate seed
    if (params.contains("seed")) {
        if (!params["seed"].is_string() && !params["seed"].is_number_integer()) {
            return {false, "Invalid 'seed' field, must be string or integer"};
        }
    }

    // Validate sampling method
    if (params.contains("sampling_method")) {
        if (!params["sampling_method"].is_string()) {
            return {false, "Invalid 'sampling_method' field, must be string"};
        }
        std::string method = params["sampling_method"];
        std::vector<std::string> validMethods = {
            "euler", "euler_a", "heun", "dpm2", "dpm++2s_a", "dpm++2m",
            "dpm++2mv2", "ipndm", "ipndm_v", "lcm", "ddim_trailing", "tcd", "default"
        };
        if (std::find(validMethods.begin(), validMethods.end(), method) == validMethods.end()) {
            return {false, "Invalid sampling method"};
        }
    }

    // Validate scheduler
    if (params.contains("scheduler")) {
        if (!params["scheduler"].is_string()) {
            return {false, "Invalid 'scheduler' field, must be string"};
        }
        std::string scheduler = params["scheduler"];
        std::vector<std::string> validSchedulers = {
            "discrete", "karras", "exponential", "ays", "gits",
            "smoothstep", "sgm_uniform", "simple", "default"
        };
        if (std::find(validSchedulers.begin(), validSchedulers.end(), scheduler) == validSchedulers.end()) {
            return {false, "Invalid scheduler"};
        }
    }

    // Validate strength
    if (params.contains("strength")) {
        if (!params["strength"].is_number()) {
            return {false, "Invalid 'strength' field, must be number"};
        }
        float strength = params["strength"];
        if (strength < 0.0f || strength > 1.0f) {
            return {false, "Strength must be between 0.0 and 1.0"};
        }
    }

    // Validate control strength
    if (params.contains("control_strength")) {
        if (!params["control_strength"].is_number()) {
            return {false, "Invalid 'control_strength' field, must be number"};
        }
        float controlStrength = params["control_strength"];
        if (controlStrength < 0.0f || controlStrength > 1.0f) {
            return {false, "Control strength must be between 0.0 and 1.0"};
        }
    }

    // Validate clip skip
    if (params.contains("clip_skip")) {
        if (!params["clip_skip"].is_number_integer()) {
            return {false, "Invalid 'clip_skip' field, must be integer"};
        }
        int clipSkip = params["clip_skip"];
        if (clipSkip < -1 || clipSkip > 12) {
            return {false, "Clip skip must be between -1 and 12"};
        }
    }

    // Validate threads
    if (params.contains("threads")) {
        if (!params["threads"].is_number_integer()) {
            return {false, "Invalid 'threads' field, must be integer"};
        }
        int threads = params["threads"];
        if (threads < -1 || threads > 32) {
            return {false, "Threads must be between -1 (auto) and 32"};
        }
    }

    return {true, ""};
}

SamplingMethod Server::parseSamplingMethod(const std::string& method) {
    if (method == "euler") return SamplingMethod::EULER;
    else if (method == "euler_a") return SamplingMethod::EULER_A;
    else if (method == "heun") return SamplingMethod::HEUN;
    else if (method == "dpm2") return SamplingMethod::DPM2;
    else if (method == "dpm++2s_a") return SamplingMethod::DPMPP2S_A;
    else if (method == "dpm++2m") return SamplingMethod::DPMPP2M;
    else if (method == "dpm++2mv2") return SamplingMethod::DPMPP2MV2;
    else if (method == "ipndm") return SamplingMethod::IPNDM;
    else if (method == "ipndm_v") return SamplingMethod::IPNDM_V;
    else if (method == "lcm") return SamplingMethod::LCM;
    else if (method == "ddim_trailing") return SamplingMethod::DDIM_TRAILING;
    else if (method == "tcd") return SamplingMethod::TCD;
    else return SamplingMethod::DEFAULT;
}

Scheduler Server::parseScheduler(const std::string& scheduler) {
    if (scheduler == "discrete") return Scheduler::DISCRETE;
    else if (scheduler == "karras") return Scheduler::KARRAS;
    else if (scheduler == "exponential") return Scheduler::EXPONENTIAL;
    else if (scheduler == "ays") return Scheduler::AYS;
    else if (scheduler == "gits") return Scheduler::GITS;
    else if (scheduler == "smoothstep") return Scheduler::SMOOTHSTEP;
    else if (scheduler == "sgm_uniform") return Scheduler::SGM_UNIFORM;
    else if (scheduler == "simple") return Scheduler::SIMPLE;
    else return Scheduler::DEFAULT;
}

std::string Server::generateRequestId() {
    std::random_device rd;
    std::mt19937 gen(rd());
    std::uniform_int_distribution<> dis(100000, 999999);
    return "req_" + std::to_string(dis(gen));
}

std::tuple<std::vector<uint8_t>, int, int, int, bool, std::string>
Server::loadImageFromInput(const std::string& input) {
    std::vector<uint8_t> imageData;
    int width = 0, height = 0, channels = 0;

    // Auto-detect input source type
    // 1. Check if input is a URL (starts with http:// or https://)
    if (Utils::startsWith(input, "http://") || Utils::startsWith(input, "https://")) {
        // Parse URL to extract host and path
        std::string url = input;
        std::string scheme, host, path;
        int port = 80;

        // Determine scheme and port
        if (Utils::startsWith(url, "https://")) {
            scheme = "https";
            port = 443;
            url = url.substr(8);  // Remove "https://"
        } else {
            scheme = "http";
            port = 80;
            url = url.substr(7);  // Remove "http://"
        }

        // Extract host and path
        size_t slashPos = url.find('/');
        if (slashPos != std::string::npos) {
            host = url.substr(0, slashPos);
            path = url.substr(slashPos);
        } else {
            host = url;
            path = "/";
        }

        // Check for custom port
        size_t colonPos = host.find(':');
        if (colonPos != std::string::npos) {
            try {
                port = std::stoi(host.substr(colonPos + 1));
                host = host.substr(0, colonPos);
            } catch (...) {
                return {imageData, 0, 0, 0, false, "Invalid port in URL"};
            }
        }

        // Download image using httplib
        try {
            httplib::Result res;

            if (scheme == "https") {
                #ifdef CPPHTTPLIB_OPENSSL_SUPPORT
                httplib::SSLClient client(host, port);
                client.set_follow_location(true);
                client.set_connection_timeout(30, 0);  // 30 seconds
                client.set_read_timeout(60, 0);  // 60 seconds
                res = client.Get(path.c_str());
                #else
                return {imageData, 0, 0, 0, false, "HTTPS not supported (OpenSSL not available)"};
                #endif
            } else {
                httplib::Client client(host, port);
                client.set_follow_location(true);
                client.set_connection_timeout(30, 0);  // 30 seconds
                client.set_read_timeout(60, 0);  // 60 seconds
                res = client.Get(path.c_str());
            }

            if (!res) {
                return {imageData, 0, 0, 0, false, "Failed to download image from URL: Connection error"};
            }

            if (res->status != 200) {
                return {imageData, 0, 0, 0, false, "Failed to download image from URL: HTTP " + std::to_string(res->status)};
            }

            // Convert response body to vector
            std::vector<uint8_t> downloadedData(res->body.begin(), res->body.end());

            // Load image from memory
            int w, h, c;
            unsigned char* pixels = stbi_load_from_memory(
                downloadedData.data(),
                downloadedData.size(),
                &w, &h, &c,
                3  // Force RGB
            );

            if (!pixels) {
                return {imageData, 0, 0, 0, false, "Failed to decode image from URL"};
            }

            width = w;
            height = h;
            channels = 3;

            size_t dataSize = width * height * channels;
            imageData.resize(dataSize);
            std::memcpy(imageData.data(), pixels, dataSize);

            stbi_image_free(pixels);

        } catch (const std::exception& e) {
            return {imageData, 0, 0, 0, false, "Failed to download image from URL: " + std::string(e.what())};
        }

    }
    // 2. Check if input is base64 encoded data URI (starts with "data:image")
    else if (Utils::startsWith(input, "data:image")) {
        // Extract base64 data after the comma
        size_t commaPos = input.find(',');
        if (commaPos == std::string::npos) {
            return {imageData, 0, 0, 0, false, "Invalid data URI format"};
        }

        std::string base64Data = input.substr(commaPos + 1);
        std::vector<uint8_t> decodedData = Utils::base64Decode(base64Data);

        // Load image from memory using stb_image
        int w, h, c;
        unsigned char* pixels = stbi_load_from_memory(
            decodedData.data(),
            decodedData.size(),
            &w, &h, &c,
            3  // Force RGB
        );

        if (!pixels) {
            return {imageData, 0, 0, 0, false, "Failed to decode image from base64 data URI"};
        }

        width = w;
        height = h;
        channels = 3;  // We forced RGB

        // Copy pixel data
        size_t dataSize = width * height * channels;
        imageData.resize(dataSize);
        std::memcpy(imageData.data(), pixels, dataSize);

        stbi_image_free(pixels);

    }
    // 3. Check if input is raw base64 (long string without slashes, likely base64)
    else if (input.length() > 100 && input.find('/') == std::string::npos && input.find('.') == std::string::npos) {
        // Likely raw base64 without data URI prefix
        std::vector<uint8_t> decodedData = Utils::base64Decode(input);

        int w, h, c;
        unsigned char* pixels = stbi_load_from_memory(
            decodedData.data(),
            decodedData.size(),
            &w, &h, &c,
            3  // Force RGB
        );

        if (!pixels) {
            return {imageData, 0, 0, 0, false, "Failed to decode image from base64"};
        }

        width = w;
        height = h;
        channels = 3;

        size_t dataSize = width * height * channels;
        imageData.resize(dataSize);
        std::memcpy(imageData.data(), pixels, dataSize);

        stbi_image_free(pixels);

    }
    // 4. Treat as local file path
    else {
        int w, h, c;
        unsigned char* pixels = stbi_load(input.c_str(), &w, &h, &c, 3);

        if (!pixels) {
            return {imageData, 0, 0, 0, false, "Failed to load image from file: " + input};
        }

        width = w;
        height = h;
        channels = 3;

        size_t dataSize = width * height * channels;
        imageData.resize(dataSize);
        std::memcpy(imageData.data(), pixels, dataSize);

        stbi_image_free(pixels);
    }

    return {imageData, width, height, channels, true, ""};
}

std::string Server::samplingMethodToString(SamplingMethod method) {
    switch (method) {
        case SamplingMethod::EULER: return "euler";
        case SamplingMethod::EULER_A: return "euler_a";
        case SamplingMethod::HEUN: return "heun";
        case SamplingMethod::DPM2: return "dpm2";
        case SamplingMethod::DPMPP2S_A: return "dpm++2s_a";
        case SamplingMethod::DPMPP2M: return "dpm++2m";
        case SamplingMethod::DPMPP2MV2: return "dpm++2mv2";
        case SamplingMethod::IPNDM: return "ipndm";
        case SamplingMethod::IPNDM_V: return "ipndm_v";
        case SamplingMethod::LCM: return "lcm";
        case SamplingMethod::DDIM_TRAILING: return "ddim_trailing";
        case SamplingMethod::TCD: return "tcd";
        default: return "default";
    }
}

std::string Server::schedulerToString(Scheduler scheduler) {
    switch (scheduler) {
        case Scheduler::DISCRETE: return "discrete";
        case Scheduler::KARRAS: return "karras";
        case Scheduler::EXPONENTIAL: return "exponential";
        case Scheduler::AYS: return "ays";
        case Scheduler::GITS: return "gits";
        case Scheduler::SMOOTHSTEP: return "smoothstep";
        case Scheduler::SGM_UNIFORM: return "sgm_uniform";
        case Scheduler::SIMPLE: return "simple";
        default: return "default";
    }
}

uint64_t Server::estimateGenerationTime(const GenerationRequest& request) {
    // Basic estimation based on parameters
    uint64_t baseTime = 1000; // 1 second base time

    // Factor in steps
    baseTime *= request.steps;

    // Factor in resolution
    double resolutionFactor = (request.width * request.height) / (512.0 * 512.0);
    baseTime = static_cast<uint64_t>(baseTime * resolutionFactor);

    // Factor in batch count
    baseTime *= request.batchCount;

    // Adjust for sampling method (some are faster than others)
    switch (request.samplingMethod) {
        case SamplingMethod::LCM:
            baseTime /= 4; // LCM is much faster
            break;
        case SamplingMethod::EULER:
        case SamplingMethod::EULER_A:
            baseTime *= 0.8; // Euler methods are faster
            break;
        case SamplingMethod::DPM2:
        case SamplingMethod::DPMPP2S_A:
            baseTime *= 1.2; // DPM methods are slower
            break;
        default:
            break;
    }

    return baseTime;
}

size_t Server::estimateMemoryUsage(const GenerationRequest& request) {
    // Basic memory estimation in bytes
    size_t baseMemory = 1024 * 1024 * 1024; // 1GB base

    // Factor in resolution
    double resolutionFactor = (request.width * request.height) / (512.0 * 512.0);
    baseMemory = static_cast<size_t>(baseMemory * resolutionFactor);

    // Factor in batch count
    baseMemory *= request.batchCount;

    // Additional memory for certain features
    if (request.diffusionFlashAttn) {
        baseMemory += 512 * 1024 * 1024; // Extra 512MB for flash attention
    }

    if (!request.controlNetPath.empty()) {
        baseMemory += 1024 * 1024 * 1024; // Extra 1GB for ControlNet
    }

    return baseMemory;
}

// Specialized generation endpoints
void Server::handleText2Img(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields for text2img
        if (!requestJson.contains("prompt") || !requestJson["prompt"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'prompt' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Validate all parameters
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);
        if (!isValid) {
            sendErrorResponse(res, errorMessage, 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Check if any model is loaded
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Get currently loaded checkpoint model
        auto allModels = m_modelManager->getAllModels();
        std::string loadedModelName;

        for (const auto& [modelName, modelInfo] : allModels) {
            if (modelInfo.type == ModelType::CHECKPOINT && modelInfo.isLoaded) {
                loadedModelName = modelName;
                break;
            }
        }

        if (loadedModelName.empty()) {
            sendErrorResponse(res, "No checkpoint model loaded. Please load a checkpoint model first using POST /api/models/{hash}/load", 400, "NO_CHECKPOINT_LOADED", requestId);
            return;
        }

        // Create generation request specifically for text2img
        GenerationRequest genRequest;
        genRequest.id = requestId;
        genRequest.modelName = loadedModelName;  // Use the currently loaded model
        genRequest.prompt = requestJson["prompt"];
        genRequest.negativePrompt = requestJson.value("negative_prompt", "");
        genRequest.width = requestJson.value("width", 512);
        genRequest.height = requestJson.value("height", 512);
        genRequest.batchCount = requestJson.value("batch_count", 1);
        genRequest.steps = requestJson.value("steps", 20);
        genRequest.cfgScale = requestJson.value("cfg_scale", 7.5f);
        genRequest.seed = requestJson.value("seed", "random");

        // Parse optional parameters
        if (requestJson.contains("sampling_method")) {
            genRequest.samplingMethod = parseSamplingMethod(requestJson["sampling_method"]);
        }
        if (requestJson.contains("scheduler")) {
            genRequest.scheduler = parseScheduler(requestJson["scheduler"]);
        }

        // Set text2img specific defaults
        genRequest.strength = 1.0f; // Full strength for text2img

        // Optional VAE model
        if (requestJson.contains("vae_model") && requestJson["vae_model"].is_string()) {
            std::string vaeModelId = requestJson["vae_model"];
            if (!vaeModelId.empty()) {
                auto vaeInfo = m_modelManager->getModelInfo(vaeModelId);
                if (!vaeInfo.name.empty() && vaeInfo.type == ModelType::VAE) {
                    genRequest.vaePath = vaeInfo.path;
                } else {
                    sendErrorResponse(res, "VAE model not found or invalid: " + vaeModelId, 400, "INVALID_VAE_MODEL", requestId);
                    return;
                }
            }
        }

        // Optional TAESD model
        if (requestJson.contains("taesd_model") && requestJson["taesd_model"].is_string()) {
            std::string taesdModelId = requestJson["taesd_model"];
            if (!taesdModelId.empty()) {
                auto taesdInfo = m_modelManager->getModelInfo(taesdModelId);
                if (!taesdInfo.name.empty() && taesdInfo.type == ModelType::TAESD) {
                    genRequest.taesdPath = taesdInfo.path;
                } else {
                    sendErrorResponse(res, "TAESD model not found or invalid: " + taesdModelId, 400, "INVALID_TAESD_MODEL", requestId);
                    return;
                }
            }
        }

        // Enqueue request
        auto future = m_generationQueue->enqueueRequest(genRequest);

        nlohmann::json params = {
            {"prompt", genRequest.prompt},
            {"negative_prompt", genRequest.negativePrompt},
            {"model", genRequest.modelName},
            {"width", genRequest.width},
            {"height", genRequest.height},
            {"batch_count", genRequest.batchCount},
            {"steps", genRequest.steps},
            {"cfg_scale", genRequest.cfgScale},
            {"seed", genRequest.seed},
            {"sampling_method", samplingMethodToString(genRequest.samplingMethod)},
            {"scheduler", schedulerToString(genRequest.scheduler)}
        };

        // Add VAE/TAESD if specified
        if (!genRequest.vaePath.empty()) {
            params["vae_model"] = requestJson.value("vae_model", "");
        }
        if (!genRequest.taesdPath.empty()) {
            params["taesd_model"] = requestJson.value("taesd_model", "");
        }

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Text-to-image generation request queued successfully"},
            {"queue_position", m_generationQueue->getQueueSize()},
            {"estimated_time_seconds", estimateGenerationTime(genRequest) / 1000},
            {"estimated_memory_mb", estimateMemoryUsage(genRequest) / (1024 * 1024)},
            {"type", "text2img"},
            {"parameters", params}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Text-to-image request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleImg2Img(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields for img2img
        if (!requestJson.contains("prompt") || !requestJson["prompt"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'prompt' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }
        if (!requestJson.contains("init_image") || !requestJson["init_image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'init_image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Validate all parameters
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);
        if (!isValid) {
            sendErrorResponse(res, errorMessage, 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Check if any model is loaded
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Get currently loaded checkpoint model
        auto allModels = m_modelManager->getAllModels();
        std::string loadedModelName;

        for (const auto& [modelName, modelInfo] : allModels) {
            if (modelInfo.type == ModelType::CHECKPOINT && modelInfo.isLoaded) {
                loadedModelName = modelName;
                break;
            }
        }

        if (loadedModelName.empty()) {
            sendErrorResponse(res, "No checkpoint model loaded. Please load a checkpoint model first using POST /api/models/{hash}/load", 400, "NO_CHECKPOINT_LOADED", requestId);
            return;
        }

        // Load the init image
        std::string initImageInput = requestJson["init_image"];
        auto [imageData, imgWidth, imgHeight, imgChannels, success, loadError] = loadImageFromInput(initImageInput);

        if (!success) {
            sendErrorResponse(res, "Failed to load init image: " + loadError, 400, "IMAGE_LOAD_ERROR", requestId);
            return;
        }

        // Create generation request specifically for img2img
        GenerationRequest genRequest;
        genRequest.id = requestId;
        genRequest.requestType = GenerationRequest::RequestType::IMG2IMG;
        genRequest.modelName = loadedModelName;  // Use the currently loaded model
        genRequest.prompt = requestJson["prompt"];
        genRequest.negativePrompt = requestJson.value("negative_prompt", "");
        genRequest.width = requestJson.value("width", imgWidth);  // Default to input image dimensions
        genRequest.height = requestJson.value("height", imgHeight);
        genRequest.batchCount = requestJson.value("batch_count", 1);
        genRequest.steps = requestJson.value("steps", 20);
        genRequest.cfgScale = requestJson.value("cfg_scale", 7.5f);
        genRequest.seed = requestJson.value("seed", "random");
        genRequest.strength = requestJson.value("strength", 0.75f);

        // Set init image data
        genRequest.initImageData = imageData;
        genRequest.initImageWidth = imgWidth;
        genRequest.initImageHeight = imgHeight;
        genRequest.initImageChannels = imgChannels;

        // Parse optional parameters
        if (requestJson.contains("sampling_method")) {
            genRequest.samplingMethod = parseSamplingMethod(requestJson["sampling_method"]);
        }
        if (requestJson.contains("scheduler")) {
            genRequest.scheduler = parseScheduler(requestJson["scheduler"]);
        }

        // Optional VAE model
        if (requestJson.contains("vae_model") && requestJson["vae_model"].is_string()) {
            std::string vaeModelId = requestJson["vae_model"];
            if (!vaeModelId.empty()) {
                auto vaeInfo = m_modelManager->getModelInfo(vaeModelId);
                if (!vaeInfo.name.empty() && vaeInfo.type == ModelType::VAE) {
                    genRequest.vaePath = vaeInfo.path;
                } else {
                    sendErrorResponse(res, "VAE model not found or invalid: " + vaeModelId, 400, "INVALID_VAE_MODEL", requestId);
                    return;
                }
            }
        }

        // Optional TAESD model
        if (requestJson.contains("taesd_model") && requestJson["taesd_model"].is_string()) {
            std::string taesdModelId = requestJson["taesd_model"];
            if (!taesdModelId.empty()) {
                auto taesdInfo = m_modelManager->getModelInfo(taesdModelId);
                if (!taesdInfo.name.empty() && taesdInfo.type == ModelType::TAESD) {
                    genRequest.taesdPath = taesdInfo.path;
                } else {
                    sendErrorResponse(res, "TAESD model not found or invalid: " + taesdModelId, 400, "INVALID_TAESD_MODEL", requestId);
                    return;
                }
            }
        }

        // Enqueue request
        auto future = m_generationQueue->enqueueRequest(genRequest);

        nlohmann::json params = {
            {"prompt", genRequest.prompt},
            {"negative_prompt", genRequest.negativePrompt},
            {"init_image", requestJson["init_image"]},
            {"model", genRequest.modelName},
            {"width", genRequest.width},
            {"height", genRequest.height},
            {"batch_count", genRequest.batchCount},
            {"steps", genRequest.steps},
            {"cfg_scale", genRequest.cfgScale},
            {"seed", genRequest.seed},
            {"strength", genRequest.strength},
            {"sampling_method", samplingMethodToString(genRequest.samplingMethod)},
            {"scheduler", schedulerToString(genRequest.scheduler)}
        };

        // Add VAE/TAESD if specified
        if (!genRequest.vaePath.empty()) {
            params["vae_model"] = requestJson.value("vae_model", "");
        }
        if (!genRequest.taesdPath.empty()) {
            params["taesd_model"] = requestJson.value("taesd_model", "");
        }

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Image-to-image generation request queued successfully"},
            {"queue_position", m_generationQueue->getQueueSize()},
            {"estimated_time_seconds", estimateGenerationTime(genRequest) / 1000},
            {"estimated_memory_mb", estimateMemoryUsage(genRequest) / (1024 * 1024)},
            {"type", "img2img"},
            {"parameters", params}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Image-to-image request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleControlNet(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields for ControlNet
        if (!requestJson.contains("prompt") || !requestJson["prompt"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'prompt' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }
        if (!requestJson.contains("control_image") || !requestJson["control_image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'control_image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Validate all parameters
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);
        if (!isValid) {
            sendErrorResponse(res, errorMessage, 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Check if any model is loaded
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Get currently loaded checkpoint model
        auto allModels = m_modelManager->getAllModels();
        std::string loadedModelName;

        for (const auto& [modelName, modelInfo] : allModels) {
            if (modelInfo.type == ModelType::CHECKPOINT && modelInfo.isLoaded) {
                loadedModelName = modelName;
                break;
            }
        }

        if (loadedModelName.empty()) {
            sendErrorResponse(res, "No checkpoint model loaded. Please load a checkpoint model first using POST /api/models/{hash}/load", 400, "NO_CHECKPOINT_LOADED", requestId);
            return;
        }

        // Create generation request specifically for ControlNet
        GenerationRequest genRequest;
        genRequest.id = requestId;
        genRequest.modelName = loadedModelName;  // Use the currently loaded model
        genRequest.prompt = requestJson["prompt"];
        genRequest.negativePrompt = requestJson.value("negative_prompt", "");
        genRequest.width = requestJson.value("width", 512);
        genRequest.height = requestJson.value("height", 512);
        genRequest.batchCount = requestJson.value("batch_count", 1);
        genRequest.steps = requestJson.value("steps", 20);
        genRequest.cfgScale = requestJson.value("cfg_scale", 7.5f);
        genRequest.seed = requestJson.value("seed", "random");
        genRequest.controlStrength = requestJson.value("control_strength", 0.9f);
        genRequest.controlNetPath = requestJson.value("control_net_model", "");

        // Parse optional parameters
        if (requestJson.contains("sampling_method")) {
            genRequest.samplingMethod = parseSamplingMethod(requestJson["sampling_method"]);
        }
        if (requestJson.contains("scheduler")) {
            genRequest.scheduler = parseScheduler(requestJson["scheduler"]);
        }

        // Optional VAE model
        if (requestJson.contains("vae_model") && requestJson["vae_model"].is_string()) {
            std::string vaeModelId = requestJson["vae_model"];
            if (!vaeModelId.empty()) {
                auto vaeInfo = m_modelManager->getModelInfo(vaeModelId);
                if (!vaeInfo.name.empty() && vaeInfo.type == ModelType::VAE) {
                    genRequest.vaePath = vaeInfo.path;
                } else {
                    sendErrorResponse(res, "VAE model not found or invalid: " + vaeModelId, 400, "INVALID_VAE_MODEL", requestId);
                    return;
                }
            }
        }

        // Optional TAESD model
        if (requestJson.contains("taesd_model") && requestJson["taesd_model"].is_string()) {
            std::string taesdModelId = requestJson["taesd_model"];
            if (!taesdModelId.empty()) {
                auto taesdInfo = m_modelManager->getModelInfo(taesdModelId);
                if (!taesdInfo.name.empty() && taesdInfo.type == ModelType::TAESD) {
                    genRequest.taesdPath = taesdInfo.path;
                } else {
                    sendErrorResponse(res, "TAESD model not found or invalid: " + taesdModelId, 400, "INVALID_TAESD_MODEL", requestId);
                    return;
                }
            }
        }

        // Store control image path (would be handled in actual implementation)
        genRequest.outputPath = requestJson.value("control_image", "");

        // Enqueue request
        auto future = m_generationQueue->enqueueRequest(genRequest);

        nlohmann::json params = {
            {"prompt", genRequest.prompt},
            {"negative_prompt", genRequest.negativePrompt},
            {"control_image", requestJson["control_image"]},
            {"control_net_model", genRequest.controlNetPath},
            {"model", genRequest.modelName},
            {"width", genRequest.width},
            {"height", genRequest.height},
            {"batch_count", genRequest.batchCount},
            {"steps", genRequest.steps},
            {"cfg_scale", genRequest.cfgScale},
            {"seed", genRequest.seed},
            {"control_strength", genRequest.controlStrength},
            {"sampling_method", samplingMethodToString(genRequest.samplingMethod)},
            {"scheduler", schedulerToString(genRequest.scheduler)}
        };

        // Add VAE/TAESD if specified
        if (!genRequest.vaePath.empty()) {
            params["vae_model"] = requestJson.value("vae_model", "");
        }
        if (!genRequest.taesdPath.empty()) {
            params["taesd_model"] = requestJson.value("taesd_model", "");
        }

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "ControlNet generation request queued successfully"},
            {"queue_position", m_generationQueue->getQueueSize()},
            {"estimated_time_seconds", estimateGenerationTime(genRequest) / 1000},
            {"estimated_memory_mb", estimateMemoryUsage(genRequest) / (1024 * 1024)},
            {"type", "controlnet"},
            {"parameters", params}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("ControlNet request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleUpscale(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields for upscaler
        if (!requestJson.contains("image") || !requestJson["image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }
        if (!requestJson.contains("esrgan_model") || !requestJson["esrgan_model"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'esrgan_model' field (model hash or name)", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Check if model manager is available
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Get the ESRGAN/upscaler model
        std::string esrganModelId = requestJson["esrgan_model"];
        auto modelInfo = m_modelManager->getModelInfo(esrganModelId);

        if (modelInfo.name.empty()) {
            sendErrorResponse(res, "ESRGAN model not found: " + esrganModelId, 404, "MODEL_NOT_FOUND", requestId);
            return;
        }

        if (modelInfo.type != ModelType::ESRGAN && modelInfo.type != ModelType::UPSCALER) {
            sendErrorResponse(res, "Model is not an ESRGAN/upscaler model", 400, "INVALID_MODEL_TYPE", requestId);
            return;
        }

        // Load the input image
        std::string imageInput = requestJson["image"];
        auto [imageData, imgWidth, imgHeight, imgChannels, success, loadError] = loadImageFromInput(imageInput);

        if (!success) {
            sendErrorResponse(res, "Failed to load image: " + loadError, 400, "IMAGE_LOAD_ERROR", requestId);
            return;
        }

        // Create upscaler request
        GenerationRequest genRequest;
        genRequest.id = requestId;
        genRequest.requestType = GenerationRequest::RequestType::UPSCALER;
        genRequest.esrganPath = modelInfo.path;
        genRequest.upscaleFactor = requestJson.value("upscale_factor", 4);
        genRequest.nThreads = requestJson.value("threads", -1);
        genRequest.offloadParamsToCpu = requestJson.value("offload_to_cpu", false);
        genRequest.diffusionConvDirect = requestJson.value("direct", false);

        // Set input image data
        genRequest.initImageData = imageData;
        genRequest.initImageWidth = imgWidth;
        genRequest.initImageHeight = imgHeight;
        genRequest.initImageChannels = imgChannels;

        // Enqueue request
        auto future = m_generationQueue->enqueueRequest(genRequest);

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Upscale request queued successfully"},
            {"queue_position", m_generationQueue->getQueueSize()},
            {"type", "upscale"},
            {"parameters", {
                {"esrgan_model", esrganModelId},
                {"upscale_factor", genRequest.upscaleFactor},
                {"input_width", imgWidth},
                {"input_height", imgHeight},
                {"output_width", imgWidth * genRequest.upscaleFactor},
                {"output_height", imgHeight * genRequest.upscaleFactor}
            }}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Upscale request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleInpainting(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue) {
            sendErrorResponse(res, "Generation queue not available", 500, "QUEUE_UNAVAILABLE", requestId);
            return;
        }

        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate required fields for inpainting
        if (!requestJson.contains("prompt") || !requestJson["prompt"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'prompt' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }
        if (!requestJson.contains("source_image") || !requestJson["source_image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'source_image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }
        if (!requestJson.contains("mask_image") || !requestJson["mask_image"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'mask_image' field", 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Validate all parameters
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);
        if (!isValid) {
            sendErrorResponse(res, errorMessage, 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Check if any model is loaded
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Get currently loaded checkpoint model
        auto allModels = m_modelManager->getAllModels();
        std::string loadedModelName;

        for (const auto& [modelName, modelInfo] : allModels) {
            if (modelInfo.type == ModelType::CHECKPOINT && modelInfo.isLoaded) {
                loadedModelName = modelName;
                break;
            }
        }

        if (loadedModelName.empty()) {
            sendErrorResponse(res, "No checkpoint model loaded. Please load a checkpoint model first using POST /api/models/{hash}/load", 400, "NO_CHECKPOINT_LOADED", requestId);
            return;
        }

        // Load the source image
        std::string sourceImageInput = requestJson["source_image"];
        auto [sourceImageData, sourceImgWidth, sourceImgHeight, sourceImgChannels, sourceSuccess, sourceLoadError] = loadImageFromInput(sourceImageInput);

        if (!sourceSuccess) {
            sendErrorResponse(res, "Failed to load source image: " + sourceLoadError, 400, "IMAGE_LOAD_ERROR", requestId);
            return;
        }

        // Load the mask image
        std::string maskImageInput = requestJson["mask_image"];
        auto [maskImageData, maskImgWidth, maskImgHeight, maskImgChannels, maskSuccess, maskLoadError] = loadImageFromInput(maskImageInput);

        if (!maskSuccess) {
            sendErrorResponse(res, "Failed to load mask image: " + maskLoadError, 400, "MASK_LOAD_ERROR", requestId);
            return;
        }

        // Validate that source and mask images have compatible dimensions
        if (sourceImgWidth != maskImgWidth || sourceImgHeight != maskImgHeight) {
            sendErrorResponse(res, "Source and mask images must have the same dimensions", 400, "DIMENSION_MISMATCH", requestId);
            return;
        }

        // Create generation request specifically for inpainting
        GenerationRequest genRequest;
        genRequest.id = requestId;
        genRequest.requestType = GenerationRequest::RequestType::INPAINTING;
        genRequest.modelName = loadedModelName;  // Use the currently loaded model
        genRequest.prompt = requestJson["prompt"];
        genRequest.negativePrompt = requestJson.value("negative_prompt", "");
        genRequest.width = requestJson.value("width", sourceImgWidth);  // Default to input image dimensions
        genRequest.height = requestJson.value("height", sourceImgHeight);
        genRequest.batchCount = requestJson.value("batch_count", 1);
        genRequest.steps = requestJson.value("steps", 20);
        genRequest.cfgScale = requestJson.value("cfg_scale", 7.5f);
        genRequest.seed = requestJson.value("seed", "random");
        genRequest.strength = requestJson.value("strength", 0.75f);

        // Set source image data
        genRequest.initImageData = sourceImageData;
        genRequest.initImageWidth = sourceImgWidth;
        genRequest.initImageHeight = sourceImgHeight;
        genRequest.initImageChannels = sourceImgChannels;

        // Set mask image data
        genRequest.maskImageData = maskImageData;
        genRequest.maskImageWidth = maskImgWidth;
        genRequest.maskImageHeight = maskImgHeight;
        genRequest.maskImageChannels = maskImgChannels;

        // Parse optional parameters
        if (requestJson.contains("sampling_method")) {
            genRequest.samplingMethod = parseSamplingMethod(requestJson["sampling_method"]);
        }
        if (requestJson.contains("scheduler")) {
            genRequest.scheduler = parseScheduler(requestJson["scheduler"]);
        }

        // Optional VAE model
        if (requestJson.contains("vae_model") && requestJson["vae_model"].is_string()) {
            std::string vaeModelId = requestJson["vae_model"];
            if (!vaeModelId.empty()) {
                auto vaeInfo = m_modelManager->getModelInfo(vaeModelId);
                if (!vaeInfo.name.empty() && vaeInfo.type == ModelType::VAE) {
                    genRequest.vaePath = vaeInfo.path;
                } else {
                    sendErrorResponse(res, "VAE model not found or invalid: " + vaeModelId, 400, "INVALID_VAE_MODEL", requestId);
                    return;
                }
            }
        }

        // Optional TAESD model
        if (requestJson.contains("taesd_model") && requestJson["taesd_model"].is_string()) {
            std::string taesdModelId = requestJson["taesd_model"];
            if (!taesdModelId.empty()) {
                auto taesdInfo = m_modelManager->getModelInfo(taesdModelId);
                if (!taesdInfo.name.empty() && taesdInfo.type == ModelType::TAESD) {
                    genRequest.taesdPath = taesdInfo.path;
                } else {
                    sendErrorResponse(res, "TAESD model not found or invalid: " + taesdModelId, 400, "INVALID_TAESD_MODEL", requestId);
                    return;
                }
            }
        }

        // Enqueue request
        auto future = m_generationQueue->enqueueRequest(genRequest);

        nlohmann::json params = {
            {"prompt", genRequest.prompt},
            {"negative_prompt", genRequest.negativePrompt},
            {"source_image", requestJson["source_image"]},
            {"mask_image", requestJson["mask_image"]},
            {"model", genRequest.modelName},
            {"width", genRequest.width},
            {"height", genRequest.height},
            {"batch_count", genRequest.batchCount},
            {"steps", genRequest.steps},
            {"cfg_scale", genRequest.cfgScale},
            {"seed", genRequest.seed},
            {"strength", genRequest.strength},
            {"sampling_method", samplingMethodToString(genRequest.samplingMethod)},
            {"scheduler", schedulerToString(genRequest.scheduler)}
        };

        // Add VAE/TAESD if specified
        if (!genRequest.vaePath.empty()) {
            params["vae_model"] = requestJson.value("vae_model", "");
        }
        if (!genRequest.taesdPath.empty()) {
            params["taesd_model"] = requestJson.value("taesd_model", "");
        }

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Inpainting generation request queued successfully"},
            {"queue_position", m_generationQueue->getQueueSize()},
            {"estimated_time_seconds", estimateGenerationTime(genRequest) / 1000},
            {"estimated_memory_mb", estimateMemoryUsage(genRequest) / (1024 * 1024)},
            {"type", "inpainting"},
            {"parameters", params}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Inpainting request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

// Utility endpoints
void Server::handleSamplers(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json samplers = {
            {"samplers", {
                {
                    {"name", "euler"},
                    {"description", "Euler sampler - fast and simple"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "euler_a"},
                    {"description", "Euler ancestral sampler - adds randomness"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "heun"},
                    {"description", "Heun sampler - more accurate but slower"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "dpm2"},
                    {"description", "DPM2 sampler - second-order DPM"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "dpm++2s_a"},
                    {"description", "DPM++ 2s ancestral sampler"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "dpm++2m"},
                    {"description", "DPM++ 2m sampler - multistep"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "dpm++2mv2"},
                    {"description", "DPM++ 2m v2 sampler - improved multistep"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "ipndm"},
                    {"description", "IPNDM sampler - improved noise prediction"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "ipndm_v"},
                    {"description", "IPNDM v sampler - variant of IPNDM"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "lcm"},
                    {"description", "LCM sampler - Latent Consistency Model, very fast"},
                    {"recommended_steps", 4}
                },
                {
                    {"name", "ddim_trailing"},
                    {"description", "DDIM trailing sampler - deterministic"},
                    {"recommended_steps", 20}
                },
                {
                    {"name", "tcd"},
                    {"description", "TCD sampler - Trajectory Consistency Distillation"},
                    {"recommended_steps", 8}
                },
                {
                    {"name", "default"},
                    {"description", "Use model's default sampler"},
                    {"recommended_steps", 20}
                }
            }}
        };

        sendJsonResponse(res, samplers);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get samplers: ") + e.what(), 500);
    }
}

void Server::handleSchedulers(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json schedulers = {
            {"schedulers", {
                {
                    {"name", "discrete"},
                    {"description", "Discrete scheduler - standard noise schedule"}
                },
                {
                    {"name", "karras"},
                    {"description", "Karras scheduler - improved noise schedule"}
                },
                {
                    {"name", "exponential"},
                    {"description", "Exponential scheduler - exponential noise decay"}
                },
                {
                    {"name", "ays"},
                    {"description", "AYS scheduler - Adaptive Your Scheduler"}
                },
                {
                    {"name", "gits"},
                    {"description", "GITS scheduler - Generalized Iterative Time Steps"}
                },
                {
                    {"name", "smoothstep"},
                    {"description", "Smoothstep scheduler - smooth transition function"}
                },
                {
                    {"name", "sgm_uniform"},
                    {"description", "SGM uniform scheduler - uniform noise schedule"}
                },
                {
                    {"name", "simple"},
                    {"description", "Simple scheduler - basic linear schedule"}
                },
                {
                    {"name", "default"},
                    {"description", "Use model's default scheduler"}
                }
            }}
        };

        sendJsonResponse(res, schedulers);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get schedulers: ") + e.what(), 500);
    }
}

void Server::handleParameters(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json parameters = {
            {"parameters", {
                {
                    {"name", "prompt"},
                    {"type", "string"},
                    {"required", true},
                    {"description", "Text prompt for image generation"},
                    {"min_length", 1},
                    {"max_length", 10000},
                    {"example", "a beautiful landscape with mountains"}
                },
                {
                    {"name", "negative_prompt"},
                    {"type", "string"},
                    {"required", false},
                    {"description", "Negative prompt to guide generation away from"},
                    {"min_length", 0},
                    {"max_length", 10000},
                    {"example", "blurry, low quality, distorted"}
                },
                {
                    {"name", "width"},
                    {"type", "integer"},
                    {"required", false},
                    {"description", "Image width in pixels"},
                    {"min", 64},
                    {"max", 2048},
                    {"multiple_of", 64},
                    {"default", 512}
                },
                {
                    {"name", "height"},
                    {"type", "integer"},
                    {"required", false},
                    {"description", "Image height in pixels"},
                    {"min", 64},
                    {"max", 2048},
                    {"multiple_of", 64},
                    {"default", 512}
                },
                {
                    {"name", "steps"},
                    {"type", "integer"},
                    {"required", false},
                    {"description", "Number of diffusion steps"},
                    {"min", 1},
                    {"max", 150},
                    {"default", 20}
                },
                {
                    {"name", "cfg_scale"},
                    {"type", "number"},
                    {"required", false},
                    {"description", "Classifier-Free Guidance scale"},
                    {"min", 1.0},
                    {"max", 30.0},
                    {"default", 7.5}
                },
                {
                    {"name", "seed"},
                    {"type", "string|integer"},
                    {"required", false},
                    {"description", "Seed for generation (use 'random' for random seed)"},
                    {"example", "42"}
                },
                {
                    {"name", "sampling_method"},
                    {"type", "string"},
                    {"required", false},
                    {"description", "Sampling method to use"},
                    {"enum", {"euler", "euler_a", "heun", "dpm2", "dpm++2s_a", "dpm++2m", "dpm++2mv2", "ipndm", "ipndm_v", "lcm", "ddim_trailing", "tcd", "default"}},
                    {"default", "default"}
                },
                {
                    {"name", "scheduler"},
                    {"type", "string"},
                    {"required", false},
                    {"description", "Scheduler to use"},
                    {"enum", {"discrete", "karras", "exponential", "ays", "gits", "smoothstep", "sgm_uniform", "simple", "default"}},
                    {"default", "default"}
                },
                {
                    {"name", "batch_count"},
                    {"type", "integer"},
                    {"required", false},
                    {"description", "Number of images to generate"},
                    {"min", 1},
                    {"max", 100},
                    {"default", 1}
                },
                {
                    {"name", "strength"},
                    {"type", "number"},
                    {"required", false},
                    {"description", "Strength for img2img (0.0-1.0)"},
                    {"min", 0.0},
                    {"max", 1.0},
                    {"default", 0.75}
                },
                {
                    {"name", "control_strength"},
                    {"type", "number"},
                    {"required", false},
                    {"description", "ControlNet strength (0.0-1.0)"},
                    {"min", 0.0},
                    {"max", 1.0},
                    {"default", 0.9}
                }
            }},
            {"openapi", {
                {"version", "3.0.0"},
                {"info", {
                    {"title", "Stable Diffusion REST API"},
                    {"version", "1.0.0"},
                    {"description", "Comprehensive REST API for stable-diffusion.cpp functionality"}
                }},
                {"components", {
                    {"schemas", {
                        {"GenerationRequest", {
                            {"type", "object"},
                            {"required", {"prompt"}},
                            {"properties", {
                                {"prompt", {{"type", "string"}, {"description", "Text prompt for generation"}}},
                                {"negative_prompt", {{"type", "string"}, {"description", "Negative prompt"}}},
                                {"width", {{"type", "integer"}, {"minimum", 64}, {"maximum", 2048}, {"default", 512}}},
                                {"height", {{"type", "integer"}, {"minimum", 64}, {"maximum", 2048}, {"default", 512}}},
                                {"steps", {{"type", "integer"}, {"minimum", 1}, {"maximum", 150}, {"default", 20}}},
                                {"cfg_scale", {{"type", "number"}, {"minimum", 1.0}, {"maximum", 30.0}, {"default", 7.5}}}
                            }}
                        }}
                    }}
                }}
            }}
        };

        sendJsonResponse(res, parameters);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get parameters: ") + e.what(), 500);
    }
}

void Server::handleValidate(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate parameters
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);

        nlohmann::json response = {
            {"request_id", requestId},
            {"valid", isValid},
            {"message", isValid ? "Parameters are valid" : errorMessage},
            {"errors", isValid ? nlohmann::json::array() : nlohmann::json::array({errorMessage})}
        };

        sendJsonResponse(res, response, isValid ? 200 : 400);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Validation failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleEstimate(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        // Validate parameters first
        auto [isValid, errorMessage] = validateGenerationParameters(requestJson);
        if (!isValid) {
            sendErrorResponse(res, errorMessage, 400, "INVALID_PARAMETERS", requestId);
            return;
        }

        // Create a temporary request to estimate
        GenerationRequest genRequest;
        genRequest.prompt = requestJson["prompt"];
        genRequest.width = requestJson.value("width", 512);
        genRequest.height = requestJson.value("height", 512);
        genRequest.batchCount = requestJson.value("batch_count", 1);
        genRequest.steps = requestJson.value("steps", 20);
        genRequest.diffusionFlashAttn = requestJson.value("diffusion_flash_attn", false);
        genRequest.controlNetPath = requestJson.value("control_net_path", "");

        if (requestJson.contains("sampling_method")) {
            genRequest.samplingMethod = parseSamplingMethod(requestJson["sampling_method"]);
        }

        // Calculate estimates
        uint64_t estimatedTime = estimateGenerationTime(genRequest);
        size_t estimatedMemory = estimateMemoryUsage(genRequest);

        nlohmann::json response = {
            {"request_id", requestId},
            {"estimated_time_seconds", estimatedTime / 1000},
            {"estimated_memory_mb", estimatedMemory / (1024 * 1024)},
            {"parameters", {
                {"resolution", std::to_string(genRequest.width) + "x" + std::to_string(genRequest.height)},
                {"steps", genRequest.steps},
                {"batch_count", genRequest.batchCount},
                {"sampling_method", samplingMethodToString(genRequest.samplingMethod)}
            }}
        };

        sendJsonResponse(res, response);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Estimation failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleConfig(const httplib::Request& /*req*/, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Get current configuration
        nlohmann::json config = {
            {"request_id", requestId},
            {"config", {
                {"server", {
                    {"host", m_host},
                    {"port", m_port},
                    {"max_concurrent_generations", 1}
                }},
                {"generation", {
                    {"default_width", 512},
                    {"default_height", 512},
                    {"default_steps", 20},
                    {"default_cfg_scale", 7.5},
                    {"max_batch_count", 100},
                    {"max_steps", 150},
                    {"max_resolution", 2048}
                }},
                {"rate_limiting", {
                    {"requests_per_minute", 60},
                    {"enabled", true}
                }}
            }}
        };

        sendJsonResponse(res, config);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Config operation failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleSystem(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json system = {
            {"system", {
                {"version", "1.0.0"},
                {"build", "stable-diffusion.cpp-rest"},
                {"uptime", std::chrono::duration_cast<std::chrono::seconds>(
                    std::chrono::steady_clock::now().time_since_epoch()).count()},
                {"capabilities", {
                    {"text2img", true},
                    {"img2img", true},
                    {"controlnet", true},
                    {"batch_generation", true},
                    {"parameter_validation", true},
                    {"estimation", true}
                }},
                {"supported_formats", {
                    {"input", {"png", "jpg", "jpeg", "webp"}},
                    {"output", {"png", "jpg", "jpeg", "webp"}}
                }},
                {"limits", {
                    {"max_resolution", 2048},
                    {"max_steps", 150},
                    {"max_batch_count", 100},
                    {"max_prompt_length", 10000}
                }}
            }},
            {"hardware", {
                {"cpu_threads", std::thread::hardware_concurrency()}
            }}
        };

        sendJsonResponse(res, system);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("System info failed: ") + e.what(), 500);
    }
}

void Server::handleSystemRestart(const httplib::Request& /*req*/, httplib::Response& res) {
    try {
        nlohmann::json response = {
            {"message", "Server restart initiated. The server will shut down gracefully and exit. Please use a process manager to automatically restart it."},
            {"status", "restarting"}
        };

        sendJsonResponse(res, response);

        // Schedule server stop after response is sent
        // Using a separate thread to allow the response to be sent first
        std::thread([this]() {
            std::this_thread::sleep_for(std::chrono::seconds(1));
            this->stop();
            // Exit with code 42 to signal restart intent to process manager
            std::exit(42);
        }).detach();

    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Restart failed: ") + e.what(), 500);
    }
}

// Helper methods for model management
nlohmann::json Server::getModelCapabilities(ModelType type) {
    nlohmann::json capabilities = nlohmann::json::object();

    switch (type) {
        case ModelType::CHECKPOINT:
            capabilities = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"outpainting", true},
                {"controlnet", true},
                {"lora", true},
                {"vae", true},
                {"sampling_methods", {"euler", "euler_a", "heun", "dpm2", "dpm++2s_a", "dpm++2m", "dpm++2mv2", "ipndm", "ipndm_v", "lcm", "ddim_trailing", "tcd"}},
                {"schedulers", {"discrete", "karras", "exponential", "ays", "gits", "smoothstep", "sgm_uniform", "simple"}},
                {"recommended_resolution", "512x512"},
                {"max_resolution", "2048x2048"},
                {"supports_batch", true}
            };
            break;
        case ModelType::LORA:
            capabilities = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", false},
                {"lora", true},
                {"vae", false},
                {"requires_checkpoint", true},
                {"strength_range", {0.0, 2.0}},
                {"recommended_strength", 1.0}
            };
            break;
        case ModelType::CONTROLNET:
            capabilities = {
                {"text2img", false},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", true},
                {"requires_checkpoint", true},
                {"control_modes", {"canny", "depth", "pose", "scribble", "hed", "mlsd", "normal", "seg"}},
                {"strength_range", {0.0, 1.0}},
                {"recommended_strength", 0.9}
            };
            break;
        case ModelType::VAE:
            capabilities = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"vae", true},
                {"requires_checkpoint", true},
                {"encoding", true},
                {"decoding", true},
                {"precision", {"fp16", "fp32"}}
            };
            break;
        case ModelType::EMBEDDING:
            capabilities = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"embedding", true},
                {"requires_checkpoint", true},
                {"token_count", 1},
                {"compatible_with", {"checkpoint", "lora"}}
            };
            break;
        case ModelType::TAESD:
            capabilities = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"vae", true},
                {"requires_checkpoint", true},
                {"fast_decoding", true},
                {"real_time", true},
                {"precision", {"fp16", "fp32"}}
            };
            break;
        case ModelType::ESRGAN:
            capabilities = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"upscaling", true},
                {"scale_factors", {2, 4}},
                {"models", {"ESRGAN", "RealESRGAN", "SwinIR"}},
                {"supports_alpha", false}
            };
            break;
        default:
            capabilities = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"capabilities", {}}
            };
            break;
    }

    return capabilities;
}

nlohmann::json Server::getModelTypeStatistics() {
    if (!m_modelManager) return nlohmann::json::object();

    nlohmann::json stats = nlohmann::json::object();
    auto allModels = m_modelManager->getAllModels();

    // Initialize counters for each type
    std::map<ModelType, int> typeCounts;
    std::map<ModelType, int> loadedCounts;
    std::map<ModelType, size_t> sizeByType;

    for (const auto& pair : allModels) {
        ModelType type = pair.second.type;
        typeCounts[type]++;
        if (pair.second.isLoaded) {
            loadedCounts[type]++;
        }
        sizeByType[type] += pair.second.fileSize;
    }

    // Build statistics JSON
    for (const auto& count : typeCounts) {
        std::string typeName = ModelManager::modelTypeToString(count.first);
        stats[typeName] = {
            {"total_count", count.second},
            {"loaded_count", loadedCounts[count.first]},
            {"total_size_bytes", sizeByType[count.first]},
            {"total_size_mb", sizeByType[count.first] / (1024.0 * 1024.0)},
            {"average_size_mb", count.second > 0 ? (sizeByType[count.first] / (1024.0 * 1024.0)) / count.second : 0.0}
        };
    }

    return stats;
}

// Additional helper methods for model management
nlohmann::json Server::getModelCompatibility(const ModelManager::ModelInfo& modelInfo) {
    nlohmann::json compatibility = {
        {"is_compatible", true},
        {"compatibility_score", 100},
        {"issues", nlohmann::json::array()},
        {"warnings", nlohmann::json::array()},
        {"requirements", {
            {"min_memory_mb", 1024},
            {"recommended_memory_mb", 2048},
            {"supported_formats", {"safetensors", "ckpt", "gguf"}},
            {"required_dependencies", {}}
        }}
    };

    // Check for specific compatibility issues based on model type
    if (modelInfo.type == ModelType::LORA) {
        compatibility["requirements"]["required_dependencies"] = {"checkpoint"};
    } else if (modelInfo.type == ModelType::CONTROLNET) {
        compatibility["requirements"]["required_dependencies"] = {"checkpoint"};
    } else if (modelInfo.type == ModelType::VAE) {
        compatibility["requirements"]["required_dependencies"] = {"checkpoint"};
    }

    return compatibility;
}

nlohmann::json Server::getModelRequirements(ModelType type) {
    nlohmann::json requirements = {
        {"min_memory_mb", 1024},
        {"recommended_memory_mb", 2048},
        {"min_disk_space_mb", 1024},
        {"supported_formats", {"safetensors", "ckpt", "gguf"}},
        {"required_dependencies", nlohmann::json::array()},
        {"optional_dependencies", nlohmann::json::array()},
        {"system_requirements", {
            {"cpu_cores", 4},
            {"cpu_architecture", "x86_64"},
            {"os", "Linux/Windows/macOS"},
            {"gpu_memory_mb", 2048},
            {"gpu_compute_capability", "3.5+"}
        }}
    };

    switch (type) {
        case ModelType::CHECKPOINT:
            requirements["min_memory_mb"] = 2048;
            requirements["recommended_memory_mb"] = 4096;
            requirements["min_disk_space_mb"] = 2048;
            requirements["supported_formats"] = {"safetensors", "ckpt", "gguf"};
            break;
        case ModelType::LORA:
            requirements["min_memory_mb"] = 512;
            requirements["recommended_memory_mb"] = 1024;
            requirements["min_disk_space_mb"] = 100;
            requirements["supported_formats"] = {"safetensors", "ckpt"};
            requirements["required_dependencies"] = {"checkpoint"};
            break;
        case ModelType::CONTROLNET:
            requirements["min_memory_mb"] = 1024;
            requirements["recommended_memory_mb"] = 2048;
            requirements["min_disk_space_mb"] = 500;
            requirements["supported_formats"] = {"safetensors", "pth"};
            requirements["required_dependencies"] = {"checkpoint"};
            break;
        case ModelType::VAE:
            requirements["min_memory_mb"] = 512;
            requirements["recommended_memory_mb"] = 1024;
            requirements["min_disk_space_mb"] = 200;
            requirements["supported_formats"] = {"safetensors", "pt", "ckpt", "gguf"};
            requirements["required_dependencies"] = {"checkpoint"};
            break;
        case ModelType::EMBEDDING:
            requirements["min_memory_mb"] = 64;
            requirements["recommended_memory_mb"] = 256;
            requirements["min_disk_space_mb"] = 10;
            requirements["supported_formats"] = {"safetensors", "pt"};
            requirements["required_dependencies"] = {"checkpoint"};
            break;
        case ModelType::TAESD:
            requirements["min_memory_mb"] = 256;
            requirements["recommended_memory_mb"] = 512;
            requirements["min_disk_space_mb"] = 100;
            requirements["supported_formats"] = {"safetensors", "pth", "gguf"};
            requirements["required_dependencies"] = {"checkpoint"};
            break;
        case ModelType::ESRGAN:
            requirements["min_memory_mb"] = 1024;
            requirements["recommended_memory_mb"] = 2048;
            requirements["min_disk_space_mb"] = 500;
            requirements["supported_formats"] = {"pth", "pt"};
            requirements["optional_dependencies"] = {"checkpoint"};
            break;
        default:
            break;
    }

    return requirements;
}

nlohmann::json Server::getRecommendedUsage(ModelType type) {
    nlohmann::json usage = {
        {"text2img", false},
        {"img2img", false},
        {"inpainting", false},
        {"controlnet", false},
        {"lora", false},
        {"vae", false},
        {"recommended_resolution", "512x512"},
        {"recommended_steps", 20},
        {"recommended_cfg_scale", 7.5},
        {"recommended_batch_size", 1}
    };

    switch (type) {
        case ModelType::CHECKPOINT:
            usage = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", true},
                {"lora", true},
                {"vae", true},
                {"recommended_resolution", "512x512"},
                {"recommended_steps", 20},
                {"recommended_cfg_scale", 7.5},
                {"recommended_batch_size", 1}
            };
            break;
        case ModelType::LORA:
            usage = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", false},
                {"lora", true},
                {"vae", false},
                {"recommended_strength", 1.0},
                {"recommended_usage", "Style transfer, character customization"}
            };
            break;
        case ModelType::CONTROLNET:
            usage = {
                {"text2img", false},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", true},
                {"lora", false},
                {"vae", false},
                {"recommended_strength", 0.9},
                {"recommended_usage", "Precise control over output"}
            };
            break;
        case ModelType::VAE:
            usage = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"controlnet", false},
                {"lora", false},
                {"vae", true},
                {"recommended_usage", "Improved encoding/decoding quality"}
            };
            break;
        case ModelType::EMBEDDING:
            usage = {
                {"text2img", true},
                {"img2img", true},
                {"inpainting", true},
                {"controlnet", false},
                {"lora", false},
                {"vae", false},
                {"embedding", true},
                {"recommended_usage", "Concept control, style words"}
            };
            break;
        case ModelType::TAESD:
            usage = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"controlnet", false},
                {"lora", false},
                {"vae", true},
                {"recommended_usage", "Real-time decoding"}
            };
            break;
        case ModelType::ESRGAN:
            usage = {
                {"text2img", false},
                {"img2img", false},
                {"inpainting", false},
                {"controlnet", false},
                {"lora", false},
                {"vae", false},
                {"upscaling", true},
                {"recommended_usage", "Image upscaling and quality enhancement"}
            };
            break;
        default:
            break;
    }

    return usage;
}

std::string Server::getModelTypeFromDirectoryName(const std::string& dirName) {
    if (dirName == "stable-diffusion" || dirName == "checkpoints") {
        return "checkpoint";
    } else if (dirName == "lora") {
        return "lora";
    } else if (dirName == "controlnet") {
        return "controlnet";
    } else if (dirName == "vae") {
        return "vae";
    } else if (dirName == "taesd") {
        return "taesd";
    } else if (dirName == "esrgan" || dirName == "upscaler") {
        return "esrgan";
    } else if (dirName == "embeddings" || dirName == "textual-inversion") {
        return "embedding";
    } else {
        return "unknown";
    }
}

std::string Server::getDirectoryDescription(const std::string& dirName) {
    if (dirName == "stable-diffusion" || dirName == "checkpoints") {
        return "Main stable diffusion model files";
    } else if (dirName == "lora") {
        return "LoRA adapter models for style transfer";
    } else if (dirName == "controlnet") {
        return "ControlNet models for precise control";
    } else if (dirName == "vae") {
        return "VAE models for improved encoding/decoding";
    } else if (dirName == "taesd") {
        return "TAESD models for real-time decoding";
    } else if (dirName == "esrgan" || dirName == "upscaler") {
        return "ESRGAN models for image upscaling";
    } else if (dirName == "embeddings" || dirName == "textual-inversion") {
        return "Text embeddings for concept control";
    } else {
        return "Unknown model directory";
    }
}

nlohmann::json Server::getDirectoryContents(const std::string& dirPath) {
    nlohmann::json contents = nlohmann::json::array();

    try {
        if (std::filesystem::exists(dirPath) && std::filesystem::is_directory(dirPath)) {
            for (const auto& entry : std::filesystem::directory_iterator(dirPath)) {
                if (entry.is_regular_file()) {
                    nlohmann::json file = {
                        {"name", entry.path().filename().string()},
                        {"path", entry.path().string()},
                        {"size", std::filesystem::file_size(entry.path())},
                        {"size_mb", std::filesystem::file_size(entry.path()) / (1024.0 * 1024.0)},
                        {"last_modified", std::chrono::duration_cast<std::chrono::seconds>(
                            std::filesystem::last_write_time(entry.path()).time_since_epoch()).count()}
                    };
                    contents.push_back(file);
                }
            }
        }
    } catch (const std::exception& e) {
        // Return empty array if directory access fails
    }

    return contents;
}

nlohmann::json Server::getLargestModel(const std::map<std::string, ModelManager::ModelInfo>& allModels) {
    nlohmann::json largest = nlohmann::json::object();
    size_t maxSize = 0;
    std::string largestName;

    for (const auto& pair : allModels) {
        if (pair.second.fileSize > maxSize) {
            maxSize = pair.second.fileSize;
            largestName = pair.second.name;
        }
    }

    if (!largestName.empty()) {
        largest = {
            {"name", largestName},
            {"size", maxSize},
            {"size_mb", maxSize / (1024.0 * 1024.0)},
            {"type", ModelManager::modelTypeToString(allModels.at(largestName).type)}
        };
    }

    return largest;
}

nlohmann::json Server::getSmallestModel(const std::map<std::string, ModelManager::ModelInfo>& allModels) {
    nlohmann::json smallest = nlohmann::json::object();
    size_t minSize = SIZE_MAX;
    std::string smallestName;

    for (const auto& pair : allModels) {
        if (pair.second.fileSize < minSize) {
            minSize = pair.second.fileSize;
            smallestName = pair.second.name;
        }
    }

    if (!smallestName.empty()) {
        smallest = {
            {"name", smallestName},
            {"size", minSize},
            {"size_mb", minSize / (1024.0 * 1024.0)},
            {"type", ModelManager::modelTypeToString(allModels.at(smallestName).type)}
        };
    }

    return smallest;
}

nlohmann::json Server::validateModelFile(const std::string& modelPath, const std::string& modelType) {
    nlohmann::json validation = {
        {"is_valid", false},
        {"errors", nlohmann::json::array()},
        {"warnings", nlohmann::json::array()},
        {"file_info", nlohmann::json::object()},
        {"compatibility", nlohmann::json::object()},
        {"recommendations", nlohmann::json::array()}
    };

    try {
        if (!std::filesystem::exists(modelPath)) {
            validation["errors"].push_back("File does not exist");
            return validation;
        }

        if (!std::filesystem::is_regular_file(modelPath)) {
            validation["errors"].push_back("Path is not a regular file");
            return validation;
        }

        // Check file extension
        std::string extension = std::filesystem::path(modelPath).extension().string();
        if (extension.empty()) {
            validation["errors"].push_back("Missing file extension");
            return validation;
        }

        // Remove dot and convert to lowercase
        if (extension[0] == '.') {
            extension = extension.substr(1);
        }
        std::transform(extension.begin(), extension.end(), extension.begin(), ::tolower);

        // Validate extension based on model type
        ModelType type = ModelManager::stringToModelType(modelType);
        bool validExtension = false;

        switch (type) {
            case ModelType::CHECKPOINT:
                validExtension = (extension == "safetensors" || extension == "ckpt" || extension == "gguf");
                break;
            case ModelType::LORA:
                validExtension = (extension == "safetensors" || extension == "ckpt");
                break;
            case ModelType::CONTROLNET:
                validExtension = (extension == "safetensors" || extension == "pth");
                break;
            case ModelType::VAE:
                validExtension = (extension == "safetensors" || extension == "pt" || extension == "ckpt" || extension == "gguf");
                break;
            case ModelType::EMBEDDING:
                validExtension = (extension == "safetensors" || extension == "pt");
                break;
            case ModelType::TAESD:
                validExtension = (extension == "safetensors" || extension == "pth" || extension == "gguf");
                break;
            case ModelType::ESRGAN:
                validExtension = (extension == "pth" || extension == "pt");
                break;
            default:
                break;
        }

        if (!validExtension) {
            validation["errors"].push_back("Invalid file extension for model type: " + extension);
        }

        // Check file size
        size_t fileSize = std::filesystem::file_size(modelPath);
        if (fileSize == 0) {
            validation["errors"].push_back("File is empty");
        } else if (fileSize > 8ULL * 1024 * 1024 * 1024) { // 8GB
            validation["warnings"].push_back("Very large file may cause performance issues");
        }

        // Build file info
        validation["file_info"] = {
            {"path", modelPath},
            {"size", fileSize},
            {"size_mb", fileSize / (1024.0 * 1024.0)},
            {"extension", extension},
            {"last_modified", std::chrono::duration_cast<std::chrono::seconds>(
                std::filesystem::last_write_time(modelPath).time_since_epoch()).count()}
        };

        // Check compatibility
        validation["compatibility"] = {
            {"extension_valid", validExtension},
            {"size_appropriate", fileSize <= 4ULL * 1024 * 1024 * 1024}, // 4GB
            {"recommended_format", "safetensors"}
        };

        // Add recommendations
        if (!validExtension) {
            validation["recommendations"].push_back("Convert to SafeTensors format for better security and performance");
        }

        if (fileSize > 2ULL * 1024 * 1024 * 1024) { // 2GB
            validation["recommendations"].push_back("Consider using a smaller model for better performance");
        }

        // If no errors found, mark as valid
        if (validation["errors"].empty()) {
            validation["is_valid"] = true;
        }

    } catch (const std::exception& e) {
        validation["errors"].push_back("Validation failed: " + std::string(e.what()));
    }

    return validation;
}

nlohmann::json Server::checkModelCompatibility(const ModelManager::ModelInfo& modelInfo, const std::string& systemInfo) {
    nlohmann::json compatibility = {
        {"is_compatible", true},
        {"compatibility_score", 100},
        {"issues", nlohmann::json::array()},
        {"warnings", nlohmann::json::array()},
        {"requirements", nlohmann::json::object()},
        {"recommendations", nlohmann::json::array()},
        {"system_info", nlohmann::json::object()}
    };

    // Check system compatibility
    if (systemInfo == "auto") {
        compatibility["system_info"] = {
            {"cpu_cores", std::thread::hardware_concurrency()}
        };
    }

    // Check model-specific compatibility issues
    if (modelInfo.type == ModelType::CHECKPOINT) {
        if (modelInfo.fileSize > 4ULL * 1024 * 1024 * 1024) { // 4GB
            compatibility["warnings"].push_back("Large checkpoint model may require significant memory");
            compatibility["compatibility_score"] = 80;
        }

        if (modelInfo.fileSize < 500 * 1024 * 1024) { // 500MB
            compatibility["warnings"].push_back("Small checkpoint model may have limited capabilities");
            compatibility["compatibility_score"] = 85;
        }
    } else if (modelInfo.type == ModelType::LORA) {
        if (modelInfo.fileSize > 500 * 1024 * 1024) { // 500MB
            compatibility["warnings"].push_back("Large LoRA may impact performance");
            compatibility["compatibility_score"] = 75;
        }
    }

    return compatibility;
}

nlohmann::json Server::calculateSpecificRequirements(const std::string& modelType, const std::string& resolution, const std::string& batchSize) {
    (void)modelType; // Suppress unused parameter warning

    nlohmann::json specific = {
        {"memory_requirements", nlohmann::json::object()},
        {"performance_impact", nlohmann::json::object()},
        {"quality_expectations", nlohmann::json::object()}
    };

    // Parse resolution
    int width = 512, height = 512;
    try {
        size_t xPos = resolution.find('x');
        if (xPos != std::string::npos) {
            width = std::stoi(resolution.substr(0, xPos));
            height = std::stoi(resolution.substr(xPos + 1));
        }
    } catch (...) {
        // Use defaults if parsing fails
    }

    // Parse batch size
    int batch = 1;
    try {
        batch = std::stoi(batchSize);
    } catch (...) {
        // Use default if parsing fails
    }

    // Calculate memory requirements based on resolution and batch
    size_t pixels = width * height;
    size_t baseMemory = 1024 * 1024 * 1024; // 1GB base
    size_t resolutionMemory = (pixels * 4) / (512 * 512); // Scale based on 512x512
    size_t batchMemory = (batch - 1) * baseMemory * 0.5; // Additional memory for batch

    specific["memory_requirements"] = {
        {"base_memory_mb", baseMemory / (1024 * 1024)},
        {"resolution_memory_mb", resolutionMemory / (1024 * 1024)},
        {"batch_memory_mb", batchMemory / (1024 * 1024)},
        {"total_memory_mb", (baseMemory + resolutionMemory + batchMemory) / (1024 * 1024)}
    };

    // Calculate performance impact
    double performanceFactor = 1.0;
    if (pixels > 512 * 512) {
        performanceFactor = 1.5;
    }
    if (batch > 1) {
        performanceFactor *= 1.2;
    }

    specific["performance_impact"] = {
        {"resolution_factor", pixels > 512 * 512 ? 1.5 : 1.0},
        {"batch_factor", batch > 1 ? 1.2 : 1.0},
        {"overall_factor", performanceFactor}
    };

    return specific;
}

// Enhanced model management endpoint implementations
void Server::handleModelInfo(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Extract model ID from URL path
        std::string modelId = req.matches[1].str();
        if (modelId.empty()) {
            sendErrorResponse(res, "Missing model ID", 400, "MISSING_MODEL_ID", requestId);
            return;
        }

        // Get model information
        auto modelInfo = m_modelManager->getModelInfo(modelId);
        if (modelInfo.name.empty()) {
            sendErrorResponse(res, "Model not found", 404, "MODEL_NOT_FOUND", requestId);
            return;
        }

        // Build comprehensive model information
        nlohmann::json response = {
            {"model", {
                {"name", modelInfo.name},
                {"path", modelInfo.path},
                {"type", ModelManager::modelTypeToString(modelInfo.type)},
                {"is_loaded", modelInfo.isLoaded},
                {"file_size", modelInfo.fileSize},
                {"file_size_mb", modelInfo.fileSize / (1024.0 * 1024.0)},
                {"description", modelInfo.description},
                {"metadata", modelInfo.metadata},
                {"capabilities", getModelCapabilities(modelInfo.type)},
                {"compatibility", getModelCompatibility(modelInfo)},
                {"requirements", getModelRequirements(modelInfo.type)},
                {"recommended_usage", getRecommendedUsage(modelInfo.type)},
                {"last_modified", std::chrono::duration_cast<std::chrono::seconds>(
                    modelInfo.modifiedAt.time_since_epoch()).count()}
            }},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get model info: ") + e.what(), 500, "MODEL_INFO_ERROR", requestId);
    }
}

void Server::handleLoadModelById(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Extract model ID from URL path (could be hash or name)
        std::string modelIdentifier = req.matches[1].str();
        if (modelIdentifier.empty()) {
            sendErrorResponse(res, "Missing model identifier", 400, "MISSING_MODEL_ID", requestId);
            return;
        }

        // Try to find by hash first (if it looks like a hash - 10+ hex chars)
        std::string modelId = modelIdentifier;
        if (modelIdentifier.length() >= 10 &&
            std::all_of(modelIdentifier.begin(), modelIdentifier.end(),
                       [](char c) { return std::isxdigit(c); })) {
            std::string foundName = m_modelManager->findModelByHash(modelIdentifier);
            if (!foundName.empty()) {
                modelId = foundName;
                std::cout << "Resolved hash " << modelIdentifier << " to model: " << modelId << std::endl;
            }
        }

        // Parse optional parameters from request body
        nlohmann::json requestJson;
        if (!req.body.empty()) {
            try {
                requestJson = nlohmann::json::parse(req.body);
            } catch (const nlohmann::json::parse_error& e) {
                sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
                return;
            }
        }

        // Unload previous model if one is loaded
        std::string previousModel;
        {
            std::lock_guard<std::mutex> lock(m_currentModelMutex);
            previousModel = m_currentlyLoadedModel;
        }

        if (!previousModel.empty() && previousModel != modelId) {
            std::cout << "Unloading previous model: " << previousModel << std::endl;
            m_modelManager->unloadModel(previousModel);
        }

        // Load model
        bool success = m_modelManager->loadModel(modelId);

        if (success) {
            // Update currently loaded model
            {
                std::lock_guard<std::mutex> lock(m_currentModelMutex);
                m_currentlyLoadedModel = modelId;
            }

            auto modelInfo = m_modelManager->getModelInfo(modelId);
            nlohmann::json response = {
                {"status", "success"},
                {"model", {
                    {"name", modelInfo.name},
                    {"path", modelInfo.path},
                    {"type", ModelManager::modelTypeToString(modelInfo.type)},
                    {"is_loaded", modelInfo.isLoaded}
                }},
                {"request_id", requestId}
            };

            sendJsonResponse(res, response);
        } else {
            sendErrorResponse(res, "Failed to load model", 400, "MODEL_LOAD_FAILED", requestId);
        }
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Model load failed: ") + e.what(), 500, "MODEL_LOAD_ERROR", requestId);
    }
}

void Server::handleUnloadModelById(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Extract model ID from URL path
        std::string modelId = req.matches[1].str();
        if (modelId.empty()) {
            sendErrorResponse(res, "Missing model ID", 400, "MISSING_MODEL_ID", requestId);
            return;
        }

        // Unload model
        bool success = m_modelManager->unloadModel(modelId);

        if (success) {
            // Clear currently loaded model if it matches
            {
                std::lock_guard<std::mutex> lock(m_currentModelMutex);
                if (m_currentlyLoadedModel == modelId) {
                    m_currentlyLoadedModel = "";
                }
            }

            nlohmann::json response = {
                {"status", "success"},
                {"model", {
                    {"name", modelId},
                    {"is_loaded", false}
                }},
                {"request_id", requestId}
            };

            sendJsonResponse(res, response);
        } else {
            sendErrorResponse(res, "Failed to unload model or model not found", 404, "MODEL_UNLOAD_FAILED", requestId);
        }
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Model unload failed: ") + e.what(), 500, "MODEL_UNLOAD_ERROR", requestId);
    }
}

void Server::handleModelTypes(const httplib::Request& /*req*/, httplib::Response& res) {

    std::string requestId = generateRequestId();

    try {
        nlohmann::json types = {
            {"model_types", {
                {
                    {"type", "checkpoint"},
                    {"description", "Main stable diffusion model files for text-to-image, image-to-image, and inpainting"},
                    {"extensions", {"safetensors", "ckpt", "gguf"}},
                    {"capabilities", {"text2img", "img2img", "inpainting", "controlnet", "lora", "vae"}},
                    {"recommended_for", "General purpose image generation"}
                },
                {
                    {"type", "lora"},
                    {"description", "LoRA adapter models for style transfer and character customization"},
                    {"extensions", {"safetensors", "ckpt"}},
                    {"capabilities", {"style_transfer", "character_customization"}},
                    {"requires", {"checkpoint"}},
                    {"recommended_for", "Style modification and character-specific generation"}
                },
                {
                    {"type", "controlnet"},
                    {"description", "ControlNet models for precise control over output composition"},
                    {"extensions", {"safetensors", "pth"}},
                    {"capabilities", {"precise_control", "composition_control"}},
                    {"requires", {"checkpoint"}},
                    {"recommended_for", "Precise control over image generation"}
                },
                {
                    {"type", "vae"},
                    {"description", "VAE models for improved encoding and decoding quality"},
                    {"extensions", {"safetensors", "pt", "ckpt", "gguf"}},
                    {"capabilities", {"encoding", "decoding", "quality_improvement"}},
                    {"requires", {"checkpoint"}},
                    {"recommended_for", "Improved image quality and encoding"}
                },
                {
                    {"type", "embedding"},
                    {"description", "Text embeddings for concept control and style words"},
                    {"extensions", {"safetensors", "pt"}},
                    {"capabilities", {"concept_control", "style_words"}},
                    {"requires", {"checkpoint"}},
                    {"recommended_for", "Concept control and specific styles"}
                },
                {
                    {"type", "taesd"},
                    {"description", "TAESD models for real-time decoding"},
                    {"extensions", {"safetensors", "pth", "gguf"}},
                    {"capabilities", {"real_time_decoding", "fast_preview"}},
                    {"requires", {"checkpoint"}},
                    {"recommended_for", "Real-time applications and fast previews"}
                },
                {
                    {"type", "esrgan"},
                    {"description", "ESRGAN models for image upscaling and enhancement"},
                    {"extensions", {"pth", "pt"}},
                    {"capabilities", {"upscaling", "enhancement", "quality_improvement"}},
                    {"recommended_for", "Image upscaling and quality enhancement"}
                }
            }},
            {"request_id", requestId}
        };

        sendJsonResponse(res, types);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get model types: ") + e.what(), 500, "MODEL_TYPES_ERROR", requestId);
    }
}

void Server::handleModelDirectories(const httplib::Request& /*req*/, httplib::Response& res) {

    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        std::string modelsDir = m_modelManager->getModelsDirectory();
        nlohmann::json directories = nlohmann::json::array();

        // Define expected model directories
        std::vector<std::string> modelDirs = {
            "stable-diffusion", "checkpoints", "lora", "controlnet",
            "vae", "taesd", "esrgan", "embeddings"
        };

        for (const auto& dirName : modelDirs) {
            std::string dirPath = modelsDir + "/" + dirName;
            std::string type = getModelTypeFromDirectoryName(dirName);
            std::string description = getDirectoryDescription(dirName);

            nlohmann::json dirInfo = {
                {"name", dirName},
                {"path", dirPath},
                {"type", type},
                {"description", description},
                {"exists", std::filesystem::exists(dirPath) && std::filesystem::is_directory(dirPath)},
                {"contents", getDirectoryContents(dirPath)}
            };

            directories.push_back(dirInfo);
        }

        nlohmann::json response = {
            {"models_directory", modelsDir},
            {"directories", directories},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get model directories: ") + e.what(), 500, "MODEL_DIRECTORIES_ERROR", requestId);
    }
}

void Server::handleRefreshModels(const httplib::Request& /*req*/, httplib::Response& res) {

    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Force refresh of model cache
        bool success = m_modelManager->scanModelsDirectory();

        if (success) {
            nlohmann::json response = {
                {"status", "success"},
                {"message", "Model cache refreshed successfully"},
                {"models_found", m_modelManager->getAvailableModelsCount()},
                {"models_loaded", m_modelManager->getLoadedModelsCount()},
                {"models_directory", m_modelManager->getModelsDirectory()},
                {"request_id", requestId}
            };

            sendJsonResponse(res, response);
        } else {
            sendErrorResponse(res, "Failed to refresh model cache", 500, "MODEL_REFRESH_FAILED", requestId);
        }
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Model refresh failed: ") + e.what(), 500, "MODEL_REFRESH_ERROR", requestId);
    }
}

void Server::handleHashModels(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue || !m_modelManager) {
            sendErrorResponse(res, "Services not available", 500, "SERVICE_UNAVAILABLE", requestId);
            return;
        }

        // Parse request body
        nlohmann::json requestJson;
        if (!req.body.empty()) {
            requestJson = nlohmann::json::parse(req.body);
        }

        HashRequest hashReq;
        hashReq.id = requestId;
        hashReq.forceRehash = requestJson.value("force_rehash", false);

        if (requestJson.contains("models") && requestJson["models"].is_array()) {
            for (const auto& model : requestJson["models"]) {
                hashReq.modelNames.push_back(model.get<std::string>());
            }
        }

        // Enqueue hash request
        auto future = m_generationQueue->enqueueHashRequest(hashReq);

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Hash job queued successfully"},
            {"models_to_hash", hashReq.modelNames.empty() ? "all_unhashed" : std::to_string(hashReq.modelNames.size())}
        };

        sendJsonResponse(res, response, 202);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Hash request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleConvertModel(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_generationQueue || !m_modelManager) {
            sendErrorResponse(res, "Services not available", 500, "SERVICE_UNAVAILABLE", requestId);
            return;
        }

        // Parse request body
        nlohmann::json requestJson;
        try {
            requestJson = nlohmann::json::parse(req.body);
        } catch (const nlohmann::json::parse_error& e) {
            sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
            return;
        }

        // Validate required fields
        if (!requestJson.contains("model_name")) {
            sendErrorResponse(res, "Missing required field: model_name", 400, "MISSING_FIELD", requestId);
            return;
        }

        if (!requestJson.contains("quantization_type")) {
            sendErrorResponse(res, "Missing required field: quantization_type", 400, "MISSING_FIELD", requestId);
            return;
        }

        std::string modelName = requestJson["model_name"].get<std::string>();
        std::string quantizationType = requestJson["quantization_type"].get<std::string>();

        // Validate quantization type
        const std::vector<std::string> validTypes = {"f32", "f16", "q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "q2_K", "q3_K", "q4_K"};
        if (std::find(validTypes.begin(), validTypes.end(), quantizationType) == validTypes.end()) {
            sendErrorResponse(res, "Invalid quantization_type. Valid types: f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0, q2_K, q3_K, q4_K",
                            400, "INVALID_QUANTIZATION_TYPE", requestId);
            return;
        }

        // Get model info to find the full path
        auto modelInfo = m_modelManager->getModelInfo(modelName);
        if (modelInfo.name.empty()) {
            sendErrorResponse(res, "Model not found: " + modelName, 404, "MODEL_NOT_FOUND", requestId);
            return;
        }

        // Check if model is already GGUF
        if (modelInfo.fullPath.find(".gguf") != std::string::npos) {
            sendErrorResponse(res, "Model is already in GGUF format. Cannot convert GGUF to GGUF.",
                            400, "ALREADY_GGUF", requestId);
            return;
        }

        // Build output path
        std::string outputPath = requestJson.value("output_path", "");
        if (outputPath.empty()) {
            // Generate default output path: model_name_quantization.gguf
            namespace fs = std::filesystem;
            fs::path inputPath(modelInfo.fullPath);
            std::string baseName = inputPath.stem().string();
            std::string outputDir = inputPath.parent_path().string();
            outputPath = outputDir + "/" + baseName + "_" + quantizationType + ".gguf";
        }

        // Create conversion request
        ConversionRequest convReq;
        convReq.id = requestId;
        convReq.modelName = modelName;
        convReq.modelPath = modelInfo.fullPath;
        convReq.outputPath = outputPath;
        convReq.quantizationType = quantizationType;

        // Enqueue conversion request
        auto future = m_generationQueue->enqueueConversionRequest(convReq);

        nlohmann::json response = {
            {"request_id", requestId},
            {"status", "queued"},
            {"message", "Model conversion queued successfully"},
            {"model_name", modelName},
            {"input_path", modelInfo.fullPath},
            {"output_path", outputPath},
            {"quantization_type", quantizationType}
        };

        sendJsonResponse(res, response, 202);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Conversion request failed: ") + e.what(), 500, "INTERNAL_ERROR", requestId);
    }
}

void Server::handleModelStats(const httplib::Request& /*req*/, httplib::Response& res) {

    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        auto allModels = m_modelManager->getAllModels();

        nlohmann::json response = {
            {"statistics", {
                {"total_models", allModels.size()},
                {"loaded_models", m_modelManager->getLoadedModelsCount()},
                {"available_models", m_modelManager->getAvailableModelsCount()},
                {"model_types", getModelTypeStatistics()},
                {"largest_model", getLargestModel(allModels)},
                {"smallest_model", getSmallestModel(allModels)}
            }},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Failed to get model stats: ") + e.what(), 500, "MODEL_STATS_ERROR", requestId);
    }
}

void Server::handleBatchModels(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        if (!requestJson.contains("operation") || !requestJson["operation"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'operation' field", 400, "INVALID_OPERATION", requestId);
            return;
        }

        if (!requestJson.contains("models") || !requestJson["models"].is_array()) {
            sendErrorResponse(res, "Missing or invalid 'models' field", 400, "INVALID_MODELS", requestId);
            return;
        }

        std::string operation = requestJson["operation"];
        nlohmann::json models = requestJson["models"];
        nlohmann::json results = nlohmann::json::array();

        for (const auto& model : models) {
            if (!model.is_string()) {
                results.push_back({
                    {"model", model},
                    {"success", false},
                    {"error", "Invalid model name"}
                });
                continue;
            }

            std::string modelName = model;
            bool success = false;
            std::string error = "";

            if (operation == "load") {
                success = m_modelManager->loadModel(modelName);
                if (!success) error = "Failed to load model";
            } else if (operation == "unload") {
                success = m_modelManager->unloadModel(modelName);
                if (!success) error = "Failed to unload model";
            } else {
                error = "Unsupported operation";
            }

            results.push_back({
                {"model", modelName},
                {"success", success},
                {"error", error.empty() ? nlohmann::json(nullptr) : nlohmann::json(error)}
            });
        }

        nlohmann::json response = {
            {"operation", operation},
            {"results", results},
            {"successful_count", std::count_if(results.begin(), results.end(),
                [](const nlohmann::json& result) { return result["success"].get<bool>(); })},
            {"failed_count", std::count_if(results.begin(), results.end(),
                [](const nlohmann::json& result) { return !result["success"].get<bool>(); })},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Batch operation failed: ") + e.what(), 500, "BATCH_OPERATION_ERROR", requestId);
    }
}

void Server::handleValidateModel(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        if (!requestJson.contains("model_path") || !requestJson["model_path"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'model_path' field", 400, "INVALID_MODEL_PATH", requestId);
            return;
        }

        std::string modelPath = requestJson["model_path"];
        std::string modelType = requestJson.value("model_type", "checkpoint");

        // Validate model file
        nlohmann::json validation = validateModelFile(modelPath, modelType);

        nlohmann::json response = {
            {"validation", validation},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Model validation failed: ") + e.what(), 500, "MODEL_VALIDATION_ERROR", requestId);
    }
}

void Server::handleCheckCompatibility(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        if (!m_modelManager) {
            sendErrorResponse(res, "Model manager not available", 500, "MODEL_MANAGER_UNAVAILABLE", requestId);
            return;
        }

        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        if (!requestJson.contains("model_name") || !requestJson["model_name"].is_string()) {
            sendErrorResponse(res, "Missing or invalid 'model_name' field", 400, "INVALID_MODEL_NAME", requestId);
            return;
        }

        std::string modelName = requestJson["model_name"];
        std::string systemInfo = requestJson.value("system_info", "auto");

        // Get model information
        auto modelInfo = m_modelManager->getModelInfo(modelName);
        if (modelInfo.name.empty()) {
            sendErrorResponse(res, "Model not found", 404, "MODEL_NOT_FOUND", requestId);
            return;
        }

        // Check compatibility
        nlohmann::json compatibility = checkModelCompatibility(modelInfo, systemInfo);

        nlohmann::json response = {
            {"model", modelName},
            {"compatibility", compatibility},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Compatibility check failed: ") + e.what(), 500, "COMPATIBILITY_CHECK_ERROR", requestId);
    }
}

void Server::handleModelRequirements(const httplib::Request& req, httplib::Response& res) {
    std::string requestId = generateRequestId();

    try {
        // Parse JSON request body
        nlohmann::json requestJson = nlohmann::json::parse(req.body);

        std::string modelType = requestJson.value("model_type", "checkpoint");
        std::string resolution = requestJson.value("resolution", "512x512");
        std::string batchSize = requestJson.value("batch_size", "1");

        // Calculate specific requirements
        nlohmann::json requirements = calculateSpecificRequirements(modelType, resolution, batchSize);

        // Get general requirements for model type
        ModelType type = ModelManager::stringToModelType(modelType);
        nlohmann::json generalRequirements = getModelRequirements(type);

        nlohmann::json response = {
            {"model_type", modelType},
            {"configuration", {
                {"resolution", resolution},
                {"batch_size", batchSize}
            }},
            {"specific_requirements", requirements},
            {"general_requirements", generalRequirements},
            {"request_id", requestId}
        };

        sendJsonResponse(res, response);
    } catch (const nlohmann::json::parse_error& e) {
        sendErrorResponse(res, std::string("Invalid JSON: ") + e.what(), 400, "JSON_PARSE_ERROR", requestId);
    } catch (const std::exception& e) {
        sendErrorResponse(res, std::string("Requirements calculation failed: ") + e.what(), 500, "REQUIREMENTS_ERROR", requestId);
    }
}

void Server::serverThreadFunction(const std::string& host, int port) {
    try {
        std::cout << "Server thread starting, attempting to bind to " << host << ":" << port << std::endl;

        // Check if port is available before attempting to bind
        std::cout << "Checking if port " << port << " is available..." << std::endl;

        // Try to create a test socket to check if port is in use
        int test_socket = socket(AF_INET, SOCK_STREAM, 0);
        if (test_socket >= 0) {
            // Set SO_REUSEADDR to avoid TIME_WAIT issues
            int opt = 1;
            if (setsockopt(test_socket, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
                std::cerr << "Warning: Failed to set SO_REUSEADDR on test socket: " << strerror(errno) << std::endl;
            }

            // Also set SO_REUSEPORT if available (for better concurrent binding handling)
            #ifdef SO_REUSEPORT
            int reuseport = 1;
            if (setsockopt(test_socket, SOL_SOCKET, SO_REUSEPORT, &reuseport, sizeof(reuseport)) < 0) {
                std::cerr << "Warning: Failed to set SO_REUSEPORT on test socket: " << strerror(errno) << std::endl;
            }
            #endif

            struct sockaddr_in addr;
            addr.sin_family = AF_INET;
            addr.sin_port = htons(port);
            addr.sin_addr.s_addr = INADDR_ANY;

            // Try to bind to the port
            if (bind(test_socket, (struct sockaddr*)&addr, sizeof(addr)) < 0) {
                close(test_socket);
                std::cerr << "ERROR: Port " << port << " is already in use! Cannot start server." << std::endl;
                std::cerr << "This could be due to:" << std::endl;
                std::cerr << "1. Another instance is already running on this port" << std::endl;
                std::cerr << "2. A previous instance crashed and the socket is in TIME_WAIT state" << std::endl;
                std::cerr << "3. The port is being used by another application" << std::endl;
                std::cerr << std::endl;
                std::cerr << "Solutions:" << std::endl;
                std::cerr << "- Wait 30-60 seconds for TIME_WAIT to expire (if server crashed)" << std::endl;
                std::cerr << "- Kill any existing processes: sudo lsof -ti:" << port << " | xargs kill -9" << std::endl;
                std::cerr << "- Use a different port with -p <port>" << std::endl;
                m_isRunning.store(false);
                m_startupFailed.store(true);
                return;
            }
            close(test_socket);
        }

        std::cout << "Port " << port << " is available, proceeding with server startup..." << std::endl;
        std::cout << "Calling listen()..." << std::endl;

        // We need to set m_isRunning after successful bind but before blocking
        // cpp-httplib doesn't provide a callback, so we set it optimistically
        // and clear it if listen() returns false
        m_isRunning.store(true);

        bool listenResult = m_httpServer->listen(host.c_str(), port);

        std::cout << "listen() returned: " << (listenResult ? "true" : "false") << std::endl;

        // If we reach here, server has stopped (either normally or due to error)
        m_isRunning.store(false);

        if (!listenResult) {
            std::cerr << "Server listen failed! This usually means port is in use or permission denied." << std::endl;
        }

    } catch (const std::exception& e) {
        std::cerr << "Exception in server thread: " << e.what() << std::endl;
        m_isRunning.store(false);
    }
}