ComfyUI Extension: ComfyUI-BAGEL

Authored by neverbiasu

Created

Updated

160 stars

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

Custom Nodes (0)

    README

    ComfyUI-Bagel

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

    About BAGEL

    <p align="center"> <img src="https://lf3-static.bytednsdoc.com/obj/eden-cn/nuhojubrps/banner.png" alt="BAGEL" width="480"/> </p>

    BAGEL is an open-source multimodal foundation model with 7B active parameters (14B total) that adopts a Mixture-of-Transformer-Experts (MoT) architecture. It is designed for multimodal understanding and generation tasks, outperforming top-tier open-source VLMs like Qwen2.5-VL and InternVL-2.5 on standard multimodal understanding leaderboards, and delivering text-to-image quality competitive with specialist generators such as SD3.

    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
    • Advanced Quantization Support: Multiple quantization modes (BF16, NF4, INT8) for the standard model
    • DFloat11 Quantized Model Support: Pre-quantized model requiring only ~22GB VRAM for single GPU setups

    Installation

    1. Model Selection and Download

    The ComfyUI-Bagel node supports automatic model selection with intelligent quantization:

    • ByteDance-Seed/BAGEL-7B-MoT: Standard model with multiple quantization options
      • BF16: Full precision mode (~80GB VRAM recommended for multi-GPU)
      • NF4: 4-bit quantization (~12-32GB VRAM, highly recommended for single GPU)
      • INT8: 8-bit quantization (~22-32GB VRAM, moderate compression)
    • DFloat11/BAGEL-7B-MoT-DF11: Pre-quantized model (~22GB VRAM, single 24GB GPU compatible)

    [!IMPORTANT] DFloat11 version do not need quantization again!

    Memory is automatically calculated based on your GPU specifications and selected quantization mode - no manual configuration needed!

    Models will be automatically downloaded to models/bagel/ when first selected. You can also manually download them:

    Standard Model

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

    DFloat11 Quantized Model (Recommended for single GPU)

    # Clone DFloat11 quantized model
git clone https://huggingface.co/DFloat11/BAGEL-7B-MoT-DF11 models/bagel/BAGEL-7B-MoT-DF11

# Or use huggingface_hub
python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='DFloat11/BAGEL-7B-MoT-DF11', local_dir='models/bagel/BAGEL-7B-MoT-DF11')"

    2. Install Dependencies

    Install the required dependencies:

    pip install -r requirements.txt

    For advanced quantization support (NF4/INT8 modes), also install:

    pip install bitsandbytes

    For DFloat11 quantized model support, also install:

    pip install dfloat11

    Note: bitsandbytes is required for NF4 and INT8 quantization modes on the standard ByteDance model. DFloat11 model works without additional quantization libraries.

    3. Restart ComfyUI

    Restart ComfyUI to load the new nodes.

    Workflows

    Text-to-Image Generation

    text to image workflow Generate high-quality images from text descriptions. Suitable for creative design and content generation.

    Image Editing Workflow

    image editing workflow Edit existing images based on textual descriptions, supporting local modifications and style adjustments.

    Image Understanding Workflow

    image understanding workflow Analyze and answer questions about image content, suitable for content understanding and information extraction.

    Performance Comparison

    | Metric | BAGEL-7B-MoT (Standard Model) | BAGEL-7B-MoT (DFloat11 Quantized Model) | |--------|-------------------------------|-----------------------------------------| | Model Size | 29.21 GB | 19.89 GB | | Peak GPU Memory (1024x1024 image generation) | 30.07 GB | 21.76 GB | | Generation Time (on an RTX4090 GPU) | 482.95 seconds | 154.39 seconds |

    DFloat11 Quantized Model significantly reduces VRAM requirements and speeds up generation time, making it ideal for single GPU setups.

    Related Links

    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.

    FAQ

    1. VRAM Requirements & Optimization

    With the new automatic memory management and quantization options:

    • 24GB GPU (Single): Use NF4 quantization or DFloat11 model for optimal performance
    • 12-16GB GPU: Use NF4 quantization for best results (may require some CPU offloading)
    • 8GB GPU: Limited support with NF4 + aggressive offloading
    • Multi-GPU: BF16 mode automatically distributes load across available GPUs
    • Memory is calculated automatically based on your hardware - no manual configuration needed!

    2. NameError: 'Qwen2Config' is not defined

    This issue is likely related to environment or dependency problems. You can install flash-attention and triton with the following links: v2.7.4.post1_crossos00 flash-attention-windows-wheel triton-windows-builds For more information, refer to this GitHub issue.