stable-diffusion.cpp-rest/include/stable_diffusion_wrapper.h

#ifndef STABLE_DIFFUSION_WRAPPER_H
#define STABLE_DIFFUSION_WRAPPER_H

#include <string>
#include <vector>
#include <memory>
#include <mutex>
#include <functional>
#include <cstdint>

// Include stable-diffusion.h to get the enum definitions
extern "C" {
    #include "stable-diffusion.h"
}

/**
 * @brief Wrapper class for stable-diffusion.cpp functionality
 *
 * This class provides a C++ interface to the stable-diffusion.cpp library,
 * handling model loading, image generation, and resource management.
 */
class StableDiffusionWrapper {
public:
    /**
     * @brief Generation parameters structure
     */
    struct GenerationParams {
        // Basic parameters
        std::string prompt;                ///< Text prompt for generation
        std::string negativePrompt;       ///< Negative prompt (optional)

        // Image parameters
        int width;                         ///< Image width
        int height;                        ///< Image height
        int batchCount;                    ///< Number of images to generate

        // Sampling parameters
        int steps;                         ///< Number of diffusion steps
        float cfgScale;                    ///< CFG scale
        std::string samplingMethod;        ///< Sampling method
        std::string scheduler;             ///< Scheduler

        // Seed control
        int64_t seed;                     ///< Seed for generation

        // Model paths
        std::string modelPath;             ///< Path to main model
        std::string clipLPath;             ///< Path to CLIP-L model
        std::string clipGPath;             ///< Path to CLIP-G model
        std::string vaePath;               ///< Path to VAE model
        std::string taesdPath;             ///< Path to TAESD model
        std::string controlNetPath;        ///< Path to ControlNet model
        std::string embeddingDir;          ///< Path to embeddings directory
        std::string loraModelDir;          ///< Path to LoRA model directory

        // Advanced parameters
        int clipSkip;                      ///< CLIP skip layers
        float strength;                    ///< Strength for img2img
        float controlStrength;             ///< ControlNet strength

        // Performance parameters
        int nThreads;                      ///< Number of threads (-1 for auto)
        bool offloadParamsToCpu;          ///< Offload parameters to CPU
        bool clipOnCpu;                    ///< Keep CLIP on CPU
        bool vaeOnCpu;                     ///< Keep VAE on CPU
        bool diffusionFlashAttn;           ///< Use flash attention
        bool diffusionConvDirect;          ///< Use direct convolution
        bool vaeConvDirect;                ///< Use direct VAE convolution

        // Model type
        std::string modelType;             ///< Model type (f32, f16, q4_0, etc.)

        // Verbose output
        bool verbose = false;               ///< Whether to print verbose model loading info

        // Constructor with default values
        GenerationParams()
            : width(512), height(512), batchCount(1), steps(20), cfgScale(7.5f),
              samplingMethod("euler"), scheduler("default"), seed(42),
              clipSkip(-1), strength(0.75f), controlStrength(0.9f),
              nThreads(-1), offloadParamsToCpu(false), clipOnCpu(false),
              vaeOnCpu(false), diffusionFlashAttn(false),
              diffusionConvDirect(false), vaeConvDirect(false),
              modelType("f16"), verbose(false) {}
    };

    /**
     * @brief Generated image structure
     */
    struct GeneratedImage {
        std::vector<uint8_t> data;        ///< Image data (RGB)
        int width;                        ///< Image width
        int height;                       ///< Image height
        int channels;                     ///< Number of channels (usually 3)
        int64_t seed;                     ///< Seed used for generation
        uint64_t generationTime;          ///< Time taken for generation in milliseconds
    };

    /**
     * @brief Progress callback function type
     *
     * @param step Current step
     * @param steps Total steps
     * @param time Time taken for current step
     * @param userData User data pointer
     */
    using ProgressCallback = std::function<void(int step, int steps, float time, void* userData)>;

    /**
     * @brief Construct a new Stable Diffusion Wrapper object
     */
    StableDiffusionWrapper();

    /**
     * @brief Destroy the Stable Diffusion Wrapper object
     */
    ~StableDiffusionWrapper();

    /**
     * @brief Load a stable-diffusion model
     *
     * @param modelPath Path to the model file
     * @param params Additional loading parameters
     * @return true if model was loaded successfully, false otherwise
     */
    bool loadModel(const std::string& modelPath, const GenerationParams& params = GenerationParams{});

    /**
     * @brief Unload the current model
     */
    void unloadModel();

    /**
     * @brief Check if a model is currently loaded
     *
     * @return true if a model is loaded, false otherwise
     */
    bool isModelLoaded() const;

    /**
     * @brief Generate an image from text prompt
     *
     * @param params Generation parameters
     * @param progressCallback Optional progress callback
     * @param userData User data for progress callback
     * @return std::vector<GeneratedImage> Generated images
     */
    std::vector<GeneratedImage> generateImage(
        const GenerationParams& params,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr
    );

    /**
     * @brief Generate an image from text prompt with image input (img2img)
     *
     * @param params Generation parameters
     * @param inputData Input image data
     * @param inputWidth Input image width
     * @param inputHeight Input image height
     * @param progressCallback Optional progress callback
     * @param userData User data for progress callback
     * @return std::vector<GeneratedImage> Generated images
     */
    std::vector<GeneratedImage> generateImageImg2Img(
        const GenerationParams& params,
        const std::vector<uint8_t>& inputData,
        int inputWidth,
        int inputHeight,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr
    );

    /**
     * @brief Generate an image with ControlNet
     *
     * @param params Generation parameters
     * @param controlData Control image data
     * @param controlWidth Control image width
     * @param controlHeight Control image height
     * @param progressCallback Optional progress callback
     * @param userData User data for progress callback
     * @return std::vector<GeneratedImage> Generated images
     */
    std::vector<GeneratedImage> generateImageControlNet(
        const GenerationParams& params,
        const std::vector<uint8_t>& controlData,
        int controlWidth,
        int controlHeight,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr
    );

    /**
     * @brief Generate an image with inpainting
     *
     * @param params Generation parameters
     * @param inputData Input image data
     * @param inputWidth Input image width
     * @param inputHeight Input image height
     * @param maskData Mask image data (grayscale, where white=keep, black=inpaint)
     * @param maskWidth Mask image width
     * @param maskHeight Mask image height
     * @param progressCallback Optional progress callback
     * @param userData User data for progress callback
     * @return std::vector<GeneratedImage> Generated images
     */
    std::vector<GeneratedImage> generateImageInpainting(
        const GenerationParams& params,
        const std::vector<uint8_t>& inputData,
        int inputWidth,
        int inputHeight,
        const std::vector<uint8_t>& maskData,
        int maskWidth,
        int maskHeight,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr
    );

    /**
     * @brief Upscale an image using ESRGAN
     *
     * @param esrganPath Path to ESRGAN model
     * @param inputData Input image data
     * @param inputWidth Input image width
     * @param inputHeight Input image height
     * @param inputChannels Number of channels in input image
     * @param upscaleFactor Upscale factor (usually 2 or 4)
     * @param nThreads Number of threads (-1 for auto)
     * @param offloadParamsToCpu Offload parameters to CPU
     * @param direct Use direct mode
     * @param progressCallback Optional progress callback
     * @param userData User data for progress callback
     * @return GeneratedImage Upscaled image
     */
    GeneratedImage upscaleImage(
        const std::string& esrganPath,
        const std::vector<uint8_t>& inputData,
        int inputWidth,
        int inputHeight,
        int inputChannels,
        uint32_t upscaleFactor,
        int nThreads = -1,
        bool offloadParamsToCpu = false,
        bool direct = false,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr);

    // File-based overload for upscaler input
    GeneratedImage upscaleImage(
        const std::string& esrganPath,
        const std::string& inputImagePath,
        uint32_t upscaleFactor,
        int nThreads = -1,
        bool offloadParamsToCpu = false,
        bool direct = false,
        ProgressCallback progressCallback = nullptr,
        void* userData = nullptr);

    /**
     * @brief Get the last error message
     *
     * @return std::string Last error message
     */
    std::string getLastError() const;

    /**
     * @brief Convert sampling method string to enum
     *
     * @param method Sampling method string
     * @return sample_method_t Sampling method enum
     */
    static sample_method_t stringToSamplingMethod(const std::string& method);

    /**
     * @brief Convert scheduler string to enum
     *
     * @param scheduler Scheduler string
     * @return scheduler_t Scheduler enum
     */
    static scheduler_t stringToScheduler(const std::string& scheduler);

    /**
     * @brief Convert model type string to enum
     *
     * @param type Model type string
     * @return sd_type_t Model type enum
     */
    static sd_type_t stringToModelType(const std::string& type);

private:
    class Impl;
    std::unique_ptr<Impl> pImpl; // Pimpl idiom

    // Thread safety
    mutable std::mutex wrapperMutex;
};

#endif // STABLE_DIFFUSION_WRAPPER_H