ComfyUI-Bagel

A custom node package for ComfyUI based on the BAGEL-7B-MoT multimodal model.

Features

• Text-to-Image Generation: Generate high-quality images using natural language prompts.
• Image Editing: Edit existing images based on textual descriptions.
• Image Understanding: Perform Q&A and analysis on images.
• Reasoning Process Display: Optionally display the model's reasoning process.
• Multiple Image Ratios: Supports 1:1, 4:3, 3:4, 16:9, 9:16 ratios.
• Advanced Parameter Control: Professional parameters like CFG scaling and timestep control.

Installation Steps

1. Model Download

First, download the BAGEL-7B-MoT model:

# Clone the model using git lfs (recommended)
git lfs install
git clone https://huggingface.co/ByteDance-Seed/BAGEL-7B-MoT models/BAGEL-7B-MoT

# Or use huggingface_hub
pip install huggingface_hub
python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='ByteDance-Seed/BAGEL-7B-MoT', local_dir='models/BAGEL-7B-MoT')"

2. Dependency Installation

Run the installation script:

python install.py

Or manually install the dependencies:

pip install -r requirements.txt

3. Restart ComfyUI

Restart ComfyUI to load the new nodes.

Node Descriptions

BAGEL Model Loader

The core node for loading BAGEL models.

Input Parameters:

• model_path: Path to the model (default: "models/BAGEL-7B-MoT")
• device_map: Device mapping strategy (auto/single/multi)
• offload: Whether to enable memory offloading

Output:

• BAGEL_MODEL: The loaded model object

BAGEL Text to Image

Text-to-image generation node.

Required Parameters:

• model: BAGEL model object
• prompt: Text prompt
• seed: Random seed (0 for random)
• image_ratio: Image aspect ratio
• cfg_text_scale: CFG text scaling (1.0-8.0)
• num_timesteps: Denoising steps (10-100)

Optional Parameters:

• show_thinking: Display reasoning process
• cfg_interval: CFG interval start value
• timestep_shift: Timestep offset
• cfg_renorm_min: CFG re-normalization minimum value
• cfg_renorm_type: CFG re-normalization type

Output:

• IMAGE: The generated image
• STRING: Reasoning process text (optional)

BAGEL Image Edit

Image editing node.

Required Parameters:

• model: BAGEL model object
• image: Input image
• prompt: Editing prompt
• seed: Random seed
• cfg_text_scale: CFG text scaling (1.0-8.0)
• cfg_img_scale: CFG image scaling (1.0-4.0)
• num_timesteps: Denoising steps

Output:

• IMAGE: The edited image
• STRING: Reasoning process text (optional)

BAGEL Image Understanding

Image understanding node.

Required Parameters:

• model: BAGEL model object
• image: Input image
• prompt: Question text

Optional Parameters:

• show_thinking: Display reasoning process
• do_sample: Enable sampling
• text_temperature: Text generation temperature (0.0-1.0)
• max_new_tokens: Maximum number of new tokens (64-4096)

Output:

• STRING: Answer text

Workflow Examples

Basic Text-to-Image

BAGEL Model Loader → BAGEL Text to Image → Save Image

Image Editing Workflow

Load Image → BAGEL Image Edit → Save Image
              ↑
BAGEL Model Loader

Image Understanding Workflow

Load Image → BAGEL Image Understanding → Preview Text
              ↑
BAGEL Model Loader

Complete Multimodal Workflow

BAGEL Model Loader → BAGEL Text to Image → BAGEL Image Edit → BAGEL Image Understanding
                    ↓ (Generate Image)        ↓ (Edit Image)      ↓ (Understand Image)
                 Save Image          Save Image        Preview Text

Usage Tips

Parameter Tuning

• CFG Text Scale: Controls the adherence to the text prompt; higher values mean stricter adherence.
• CFG Image Scale: (During editing) Controls the degree of preservation of the original image.
• Timestep Shift: Higher values emphasize layout, lower values emphasize details.
• CFG Renorm Type: If the generated image is blurry, try using "global".

Performance Optimization

• Use device_map="auto" to automatically allocate GPU memory.
• Enable offload=True to unload the model to disk when memory is insufficient.
• Fewer num_timesteps can speed up generation but may reduce quality.

Prompt Suggestions

• Image Generation: Describe the scene, style, color, lighting, etc., in detail.
• Image Editing: Clearly specify the parts to be modified and the desired effects.
• Image Understanding: Ask specific questions about the content, emotion, technical details, etc.

System Requirements

• GPU: Recommended 24GB+ VRAM (can use memory offloading to reduce requirements)
• RAM: 16GB+ system memory
• Storage: About 15GB of free space for model files
• Python: Version 3.8+
• CUDA: Version 11.8+ (recommended)

Troubleshooting

Common Issues

1. Model Loading Failed

   • Check if the model path is correct
   • Ensure the model files are fully downloaded
   • Check if there is enough storage space

2. Insufficient Memory

   • Enable offload=True
   • Use device_map="single" for single GPU mode
   • Reduce memory usage of other applications

3. Poor Image Quality

   • Adjust CFG parameters
   • Increase the number of timesteps
   • Optimize the prompt description

4. Import Errors

   • Reinstall dependencies: pip install -r requirements.txt
   • Check CUDA version compatibility
   • Restart ComfyUI

Log Viewing

Check the ComfyUI console output for detailed error information.

Changelog

v1.0.0

• Initial release
• Supports text-to-image generation, image editing, and image understanding
• Complete parameter control and error handling
• Memory optimization and multi-GPU support

Related Links

• BAGEL Official Paper
• BAGEL Model Homepage
• Hugging Face Model
• Online Demo
• Discord Community

License

This project is licensed under the Apache 2.0 License. Please refer to the official license terms for the use of the BAGEL model.

Contribution

Contributions are welcome! Please submit issue reports and feature requests. If you wish to contribute code, please create an issue to discuss your ideas first.