Skip to content

update.md

Latest commit

 

History

History
176 lines (143 loc) · 9.39 KB

update.md

File metadata and controls

176 lines (143 loc) · 9.39 KB

ComfyUI-JoyCaption Update Log

V2.0.2 (2025-10-27)

🧩 New Feature: Custom Model Support

  • Added custom_models.json system — users can now add their own models without modifying the core repository. #19
  • Supports both Hugging Face (hf_models) and GGUF (gguf_models) formats.
  • Fully optional — if the file is missing or empty, JoyCaption continues as normal.
  • Safe and mergeable — your custom definitions are loaded dynamically at runtime.
  • Includes example file: custom_models_example.json
  • Added detailed documentation: 📘 custom_models.md

V2.0.1 (2025-10-20)

🐛 Bug Fixes

  • Fixed Module Loading Issues: Resolved "ModuleNotFoundError: No module named 'JC'" by implementing sorted file loading in __init__.py
  • Fixed CUDA Memory Issues: Improved error handling for CUDA memory allocation failures

🔧 Code Improvements

  • Inline Utility Functions: Moved error handling and utility functions back into main modules to avoid import conflicts
  • Enhanced Error Handling: Better error messages and resource cleanup for model loading failures
  • Improved Memory Management: More robust GPU memory cleanup and error recovery

🎨 User Experience

  • Added Progress Feedback: Console output now shows processing status for GGUF models
  • Enhanced Processing Mode: Added "Auto" option to processing_mode (Auto/GPU/CPU) for better hardware detection
  • Better Tooltips: Improved parameter descriptions and user guidance

📝 Documentation

  • Fixed Installation Guide: Corrected path error in Chinese installation guide (llama_cpp_install_zh.md)
  • Updated Model Path Instructions: Added clearer guidance for model placement in README

🛠️ Technical Changes

  • Configuration Centralization: Added gguf_settings to jc_data.json for better configuration management
  • Stabilized Module Loading: Implemented deterministic file loading order to prevent import race conditions

V2.0.0 (2025-08-22)

Joycaption GGUF https://github.com/1038lab/ComfyUI-JoyCaption/blob/main/example_workflows/JoyCaption-GGUF.json

🚀 Major Features

  • GGUF Model Support: Added comprehensive support for quantized GGUF models for better performance and lower memory usage
    • New JoyCaption GGUF node for basic GGUF model usage
    • New JoyCaption GGUF (Advanced) node with full parameter control
    • Support for multiple quantization levels (Q2_K, Q3_K_S, Q3_K_M, Q3_K_L, IQ4_XS, Q4_K_S, Q4_K_M, Q5_K_S, Q5_K_M, Q6_K, Q8_0, F16)
    • Automatic model and vision projection model downloading
    • llama_cpp_install folder: Added comprehensive installation guides and scripts for llama-cpp-python

🔧 Critical Bug Fixes

  • Fixed TypeError: Resolved 'NoneType' object cannot be interpreted as an integer error with top_k parameter
  • Fixed Image Format: Updated image processing to use correct base64 data URI format for llama-cpp-python
  • Fixed Missing Variables: Resolved NameError: name 'prompt_text' is not defined in JC_GGUF class
  • Enhanced Error Handling: Improved llama-cpp-python dependency detection and fallback mechanisms

🎨 User Experience Improvements

  • Clean Console Output: Eliminated base64 image data spam in backend console
  • Enhanced Error Handling: Improved error messages and parameter validation
  • Better Performance: GGUF models provide 2-4x speed improvement with lower memory usage
  • Simplified Installation: One-click llama-cpp-python installation with automatic CUDA support

📊 Performance Optimizations

  • Memory Efficiency: GGUF models use 50-80% less memory than standard models
  • Processing Speed: Faster inference with quantized models
  • Output Suppression: Clean console output during model loading and generation
  • Optimized Loading: Improved model loading times and memory management

🛠️ Technical Improvements

  • Parameter Handling: Proper top_k parameter validation (only included when > 0)
  • Image Processing: Correct PIL to base64 conversion for vision models
  • Code Quality: Cleaned up production code, removed unnecessary comments
  • Stability: Enhanced error recovery and model management
  • Installation Automation: Streamlined llama-cpp-python installation process

📋 Model Support

  • Standard Models: Continue to support HuggingFace format models
  • GGUF Models: Comprehensive support for efficient quantized models
    • Q2_K: 3.18GB, ~4GB VRAM, Good quality, Low VRAM systems (6GB+)
    • Q3_K_S: 3.66GB, ~5GB VRAM, Good+ quality, Budget systems (8GB+)
    • Q3_K_M: 4.02GB, ~5GB VRAM, Better quality, Balanced performance
    • Q3_K_L: 4.32GB, ~6GB VRAM, Better+ quality, Good quality/size ratio
    • IQ4_XS: 4.48GB, ~6GB VRAM, Very Good quality, Recommended (8GB+)
    • Q4_K_S: 4.69GB, ~6GB VRAM, Very Good quality, Quality focused
    • Q4_K_M: 4.92GB, ~7GB VRAM, Very Good+ quality, Balanced choice
    • Q5_K_S: 5.60GB, ~7GB VRAM, Excellent quality, High quality (10GB+)
    • Q5_K_M: 5.73GB, ~8GB VRAM, Excellent+ quality, Premium quality
    • Q6_K: 6.60GB, ~8GB VRAM, Near Original quality, Maximum quality (12GB+)
    • Q8_0: 8.54GB, ~10GB VRAM, Original- quality, Full precision alternative
    • F16: 16.1GB, ~18GB VRAM, Original quality, Full precision (24GB+)

📚 Installation Resources

  • Automated Installation: llama_cpp_install/llama_cpp_install.py - One-click installation script
  • English Guide: llama_cpp_install/llama_cpp_install.md - Detailed English installation instructions
  • Chinese Guide: llama_cpp_install/llama_cpp_install_zh.md - 中文安装指南
  • Cross-Platform Support: Windows, macOS, and Linux installation guides
  • CUDA Support: Automatic GPU detection and CUDA compilation
  • Pre-compiled Wheels: Faster installation with pre-built binaries

V1.2.0 (2025-06-15)

Performance Optimizations

  • Enhanced CUDA performance for high-end GPUs
  • Optimized model loading and caching system
  • Improved memory management strategies
    • Added Global Cache mode for faster processing with sufficient VRAM
    • Enhanced Keep in Memory mode for balanced performance
    • Optimized Clear After Run mode for limited VRAM scenarios
  • Implemented efficient memory cleanup after processing
  • Added automatic memory usage monitoring

Memory Management

  • Optimized CUDA memory allocation with max_split_size_mb configuration
  • Enhanced model caching system with validation checks
  • Improved memory cleanup during model switching
  • Added automatic GPU memory management for different VRAM capacities

Code Improvements

  • Enhanced error handling and recovery mechanisms
  • Improved model loading validation
  • Optimized image processing pipeline
  • Added better tooltips and documentation for node parameters

V1.1.1 (2025-06-07)

Bug Fixes

  • Fixed CaptionTool nodes not registering in ComfyUI interface

Internationalization (i18n)

  • Added multi-language support for node interfaces
    • English (en) - Default language
    • French (fr) - Support for French interface
    • Japanese (ja) - 日本語インターフェース対応
    • Korean (ko) - 한국어 인터페이스 지원
    • Russian (ru) - Поддержка русского интерфейса
    • Chinese (zh) - 中文界面支持

V1.1.0 (2025-06-05)

Features

  • Initial release of ComfyUI-JoyCaption
  • Added JoyCaption node for image captioning
  • Integrated memory management system
  • Added caption tools for text processing

Joycaption_node

Batch Caption

Memory Management

  • Implemented efficient memory handling for large image processing
  • Added automatic memory cleanup after processing
  • Optimized memory usage during batch operations
  • Added memory usage monitoring

Caption Tools

  • Added Image Batch Path node (🖼️) for batch image loading
    • Support for sequential, reverse, and random image loading
    • Configurable batch size and start position
    • Automatic EXIF orientation correction
    • Support for jpg, jpeg, png, and webp formats
  • Added Caption Saver node (📝) for caption management

    • Flexible output path configuration

    • Custom filename support

    • Optional image copying with captions

    • Automatic file overwrite protection

    • UTF-8 encoding support

    • Batch processing capability