ComfyUI Qwen-Image DFloat11 Nodes

ComfyUI custom nodes for the DFloat11 compressed Qwen-Image model. This package provides efficient image generation with reduced memory usage through DFloat11 compression technology.

Features

• DFloat11 Qwen-Image Loader: Load the compressed Qwen-Image model with CPU offloading options
• Qwen-Image Text Encode: Process text prompts with language-specific magic prompts
• Qwen-Image Sampler: Generate images with full control over parameters
• Qwen-Image Decode: Save generated images to output directory
• Qwen-Image Aspect Ratio: Quick aspect ratio selection for common formats
• Qwen-Image Preset Sampler: Fast generation with preset configurations

Installation

1. Install Custom Nodes

Clone this repository into your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/verIdyia/ComfyUI-Qwen-Image-DF11.git

Install dependencies:

cd ComfyUI-Qwen-Image-DF11
pip install -r requirements.txt

2. Download Models

Download the complete Qwen-Image model structure with DFloat11 compressed transformer:

Required Model Structure:

ComfyUI/models/diffusion_models/
└── Qwen-Image/
    ├── model_index.json              # From Qwen/Qwen-Image
    ├── transformer/
    │   ├── config.json               # From Qwen/Qwen-Image
    │   └── diffusion_pytorch_model.safetensors  # From DFloat11/Qwen-Image-DF11
    ├── text_encoder/
    │   ├── config.json               # From Qwen/Qwen-Image
    │   ├── model.safetensors         # From Qwen/Qwen-Image
    │   └── ...
    ├── vae/
    │   ├── config.json               # From Qwen/Qwen-Image
    │   ├── diffusion_pytorch_model.safetensors  # From Qwen/Qwen-Image
    │   └── ...
    ├── tokenizer/
    │   ├── tokenizer_config.json     # From Qwen/Qwen-Image
    │   ├── tokenizer.json            # From Qwen/Qwen-Image
    │   └── ...
    └── scheduler/
        └── scheduler_config.json     # From Qwen/Qwen-Image

Installation Steps:

1. Download base model structure:

cd ComfyUI/models/diffusion_models
git lfs clone https://huggingface.co/Qwen/Qwen-Image

2. Replace transformer with DFloat11 compressed version:

cd Qwen-Image/transformer
# Backup original transformer (optional)
mv diffusion_pytorch_model.safetensors diffusion_pytorch_model.safetensors.backup
# Download DFloat11 compressed transformer
wget https://huggingface.co/DFloat11/Qwen-Image-DF11/resolve/main/diffusion_pytorch_model.safetensors

OR using Hugging Face Hub:

from huggingface_hub import hf_hub_download
hf_hub_download("DFloat11/Qwen-Image-DF11", "diffusion_pytorch_model.safetensors", 
                local_dir="ComfyUI/models/diffusion_models/Qwen-Image/transformer")

Memory Benefits:

• Original transformer: ~41GB
• DFloat11 compressed: ~28GB (32% size reduction)
• Other components remain unchanged for compatibility

3. Restart ComfyUI

Restart ComfyUI to load the new nodes.

Memory Requirements

• Without CPU Offloading: 32GB VRAM required
• With CPU Offloading: 16GB VRAM required
• Model Size: 28.42 GB (32% smaller than original BFloat16)

Usage

See the included example workflow JSON file for a complete setup example.

Model Information

This package uses the DFloat11 compressed version of Qwen-Image:

• Model: DFloat11/Qwen-Image-DF11
• Original: Qwen/Qwen-Image
• Compression: Lossless 32% size reduction
• Performance: Bit-identical outputs to original model

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

The underlying Qwen-Image model is also licensed under Apache 2.0. Please refer to the original model's license terms:

• Qwen-Image - Apache 2.0
• DFloat11/Qwen-Image-DF11 - Please check the model's license on Hugging Face