ComfyUI Extension: HYPIR ComfyUI Plugin

Authored by 11dogzi

Created

Updated

64 stars

This is a ComfyUI plugin for a/HYPIR (Harnessing Diffusion-Yielded Score Priors for Image Restoration), a state-of-the-art image restoration model based on Stable Diffusion 2.1.

Custom Nodes (0)

    README

    原项目地址https://github.com/XPixelGroup/HYPIR?tab=readme-ov-file

    HYPIR ComfyUI Plugin

    HYPIR ComfyUI 插件

    This is a ComfyUI plugin for HYPIR (Harnessing Diffusion-Yielded Score Priors for Image Restoration), a state-of-the-art image restoration model based on Stable Diffusion 2.1. 这是一个用于 HYPIR（利用扩散得分先验进行图像修复） 的 ComfyUI 插件，HYPIR 是基于 Stable Diffusion 2.1 的先进图像修复模型。

    Features

    功能特性

    • Image Restoration: Restore and enhance low-quality images using diffusion priors
    • 图像修复：利用扩散先验修复和增强低质量图像
    • Batch Processing: Process multiple images at once
    • 批量处理：一次处理多张图片
    • Advanced Controls: Fine-tune model parameters for optimal results
    • 高级控制：可微调模型参数以获得最佳效果
    • Model Management: Load and reuse HYPIR models efficiently
    • 模型管理：高效加载和复用 HYPIR 模型
    • Upscaling: Built-in upscaling capabilities (1x to 8x)
    • 放大功能：内置放大功能（1x 到 8x）

    Installation

    安装方法

    1. Install the Plugin

    1. 安装插件

    Place this folder in your ComfyUI custom_nodes directory: 将本文件夹放入 ComfyUI 的 custom_nodes 目录下：

    ComfyUI/custom_nodes/Comfyui-HYPIR/

    2. Install HYPIR Dependencies

    2. 安装 HYPIR 依赖

    Navigate to the HYPIR folder and install the required dependencies: 进入 HYPIR 文件夹并安装所需依赖：

    cd ComfyUI/custom_nodes/Comfyui-HYPIR/HYPIR
pip install -r requirements.txt

    3. Model Download (Automatic)

    3. 模型下载（自动）

    The plugin will automatically download the required models on first use: 插件首次使用时会自动下载所需模型：

    HYPIR Model

    HYPIR 模型

    The HYPIR restoration model will be downloaded to: HYPIR 修复模型将下载到：

    ComfyUI/models/HYPIR/HYPIR_sd2.pth

    Base Model (Stable Diffusion 2.1)

    基础模型（Stable Diffusion 2.1）

    The base Stable Diffusion 2.1 model will be automatically downloaded when needed to: 基础 Stable Diffusion 2.1 模型将在需要时自动下载到：

    ComfyUI/models/HYPIR/stable-diffusion-2-1-base/

    Manual Download (Optional): 手动下载（可选）：

    HYPIR Model: HYPIR 模型： If you prefer to download manually, you can get the HYPIR model from: 如果你希望手动下载，可以从以下地址获取 HYPIR 模型：

    Place the HYPIR_sd2.pth file in: 请将 HYPIR_sd2.pth 文件放在以下任一位置：

    • Plugin directory: ComfyUI/custom_nodes/Comfyui-HYPIR/
    • 插件目录：ComfyUI/custom_nodes/Comfyui-HYPIR/
    • ComfyUI models directory: ComfyUI/models/checkpoints/
    • ComfyUI 模型目录：ComfyUI/models/checkpoints/
    • Or let the plugin automatically manage it in ComfyUI/models/HYPIR/
    • 或让插件自动管理，放在 ComfyUI/models/HYPIR/

    Base Model: 基础模型： The base Stable Diffusion 2.1 model can be downloaded manually from: 基础 Stable Diffusion 2.1 模型可从以下地址手动下载：

    Place the base model in: 请将基础模型放在：

    ComfyUI/models/HYPIR/stable-diffusion-2-1-base/

    Note: The plugin will automatically check for the base model in the HYPIR directory first. If not found, it will automatically download it from HuggingFace. 注意： 插件会优先在 HYPIR 目录下查找基础模型，如未找到会自动从 HuggingFace 下载。

    Usage

    使用方法

    Advanced Image Restoration

    高级图像修复

    1. Add the HYPIR Advanced Restoration node
    2. 添加 HYPIR Advanced Restoration 节点
    3. This node provides additional control over:
    4. 此节点提供更多参数控制：
      • model_t: Model timestep (default: 200)
      • model_t：模型步数（默认200）
      • coeff_t: Coefficient timestep (default: 200)
      • coeff_t：系数步数（默认200）
      • lora_rank: LoRA rank (default: 256)
      • lora_rank：LoRA 阶数（默认256）
      • patch_size: Processing patch size (default: 512)
      • patch_size：处理块大小（默认512）

    Configuration

    配置

    You can modify the default settings in hypir_config.py: 你可以在 hypir_config.py 中修改默认设置：

    HYPIR_CONFIG = {
    "default_weight_path": "HYPIR_sd2.pth",
    "default_base_model_path": "stable-diffusion-2-1-base",
    "available_base_models": ["stable-diffusion-2-1-base"],
    "model_t": 200,
    "coeff_t": 200,
    "lora_rank": 256,
    # ... more settings
}

    Model Path Management

    模型路径管理

    The plugin includes intelligent model path management: 插件包含智能模型路径管理：

    • HYPIR Model: Automatically downloaded to ComfyUI/models/HYPIR/HYPIR_sd2.pth
    • HYPIR 模型：自动下载到 ComfyUI/models/HYPIR/HYPIR_sd2.pth
    • Base Model: Automatically downloaded to ComfyUI/models/HYPIR/stable-diffusion-2-1-base/ when needed
    • 基础模型：需要时自动下载到 ComfyUI/models/HYPIR/stable-diffusion-2-1-base/
    • Local Priority: The plugin checks for local models first before downloading
    • 本地优先：插件会优先查找本地模型
    • Automatic Download: Only downloads when models are not found locally
    • 自动下载：仅在本地未找到模型时才下载

    Tips for Best Results

    最佳效果小贴士

    1. Prompts: Use descriptive prompts that match the image content

    2. 提示词：使用与图片内容相符的描述性提示词

      • For portraits: "high quality portrait, detailed face, sharp features"
      • 人像："high quality portrait, detailed face, sharp features"
      • For landscapes: "high quality landscape, detailed scenery, sharp focus"
      • 风景："high quality landscape, detailed scenery, sharp focus"
      • For general images: "high quality, detailed, sharp, clear"
      • 通用："high quality, detailed, sharp, clear"

    3. Upscaling:

    4. 放大：

      • Use 1x for restoration without size change
      • 1x 表示仅修复不放大
      • Use 2x-4x for moderate upscaling
      • 2x-4x 适合中等放大
      • Use 8x for maximum upscaling (may be slower)
      • 8x 为最大放大（速度较慢）

    5. Parameters:

    6. 参数：

      • Higher model_t values (200-500) for stronger restoration
      • model_t 越高（200-500）修复越强
      • Higher coeff_t values (200-500) for more aggressive enhancement
      • coeff_t 越高（200-500）增强越明显
      • Higher lora_rank (256-512) for better quality (uses more memory)
      • lora_rank 越高（256-512）质量越好（占用更多内存）

    7. Memory Management:

    8. 内存管理：

      • Use smaller patch_size (256-512) if you encounter memory issues
      • 如遇内存不足可用较小的 patch_size（256-512）
      • Process images in smaller batches
      • 分批处理图片
      • Use the Model Loader node to avoid repeated model loading
      • 使用模型加载器节点避免重复加载模型

    Troubleshooting

    故障排查

    Common Issues

    常见问题

    1. Import Error: Make sure HYPIR dependencies are installed

    2. 导入错误：请确保已安装 HYPIR 依赖

      cd HYPIR
pip install -r requirements.txt

    3. Model Not Found: The plugin will automatically download missing models

    4. 模型未找到：插件会自动下载缺失的模型

      • Check internet connection for automatic download
      • 检查网络连接以便自动下载
      • HYPIR Model: Place HYPIR_sd2.pth in plugin directory or ComfyUI models directory
      • HYPIR 模型：将 HYPIR_sd2.pth 放在插件目录或 ComfyUI 模型目录
      • Base Model: Place stable-diffusion-2-1-base folder in ComfyUI/models/HYPIR/
      • 基础模型：将 stable-diffusion-2-1-base 文件夹放在 ComfyUI/models/HYPIR/
      • The plugin will automatically check and download missing models
      • 插件会自动检查并下载缺失模型

    5. CUDA Out of Memory:

    6. CUDA 内存溢出：

      • Reduce patch_size to 256
      • 将 patch_size 降至 256
      • Process smaller images
      • 处理较小的图片
      • Use CPU mode if available
      • 如可用可切换到 CPU 模式

    7. Slow Processing:

    8. 处理缓慢：

      • Use Model Loader to avoid repeated model loading
      • 使用模型加载器避免重复加载模型
      • Reduce upscale factor
      • 降低放大倍数
      • Use smaller images
      • 使用较小的图片

    Error Messages

    错误信息

    • "Error importing HYPIR": Install HYPIR dependencies
    • “Error importing HYPIR”：请安装 HYPIR 依赖
    • "Error loading model": Check model file path and permissions
    • “Error loading model”：请检查模型文件路径和权限
    • "Error during restoration": Check image format and size
    • “Error during restoration”：请检查图片格式和尺寸

    License

    许可证

    This plugin is provided under the same license as the original HYPIR project. Please refer to the HYPIR repository for license details. 本插件遵循原 HYPIR 项目的相同许可证。详情请参见 HYPIR 仓库。

    Acknowledgments

    致谢

    • Original HYPIR implementation by XPixelGroup
    • HYPIR 原始实现：XPixelGroup
    • ComfyUI community for the excellent framework
    • ComfyUI 社区提供了优秀的框架

    Support

    支持

    For issues related to this ComfyUI plugin, please check: 如遇本插件相关问题，请参考：

    1. This README for troubleshooting
    2. 本 README 故障排查部分
    3. The original HYPIR repository for model-specific issues
    4. 原 HYPIR 仓库 以获取模型相关问题
    5. ComfyUI documentation for general ComfyUI questions
    6. ComfyUI 文档以了解通用 ComfyUI 问题