https://github.com/LeanModels/ComfyUI-DFloat11
https://github.com/LeanModels/ComfyUI-DFloat11
Last synced: 7 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/LeanModels/ComfyUI-DFloat11
- Owner: LeanModels
- Created: 2025-08-26T07:59:22.000Z (7 months ago)
- Default Branch: master
- Last Pushed: 2025-08-26T08:55:44.000Z (7 months ago)
- Last Synced: 2025-08-26T10:52:46.955Z (7 months ago)
- Language: Python
- Size: 1.5 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-comfyui - **ComfyUI-DFloat11** - for-bit identical outputs to the original. Unlike quantization techniques which trade quality for size, DFloat11 is a lossless compression method, preserving model output quality fully while supporting efficient inference. (Workflows (3395) sorted by GitHub Stars)
README
# ComfyUI Plugin for DFloat11
[](https://pepy.tech/projects/dfloat11)
[](https://arxiv.org/abs/2504.11651)
[](https://huggingface.co/DFloat11)
This repository provides the **ComfyUI plugin for DFloat11 models**.
DFloat11 reduces model size by more than **30%** while producing **bit-for-bit identical outputs** to the original. Unlike quantization techniques which trade quality for size, DFloat11 is a **lossless compression method**, preserving model output quality fully while supporting efficient inference.
Currently, only **FLUX.1 models** are supported. Support for additional models is planned. Please feel free to [open an issue](https://github.com/LeanModels/ComfyUI-DFloat11/issues) and let us know which ones you'd like to see next.
---
## Features
* 🚀 **Fully Lossless** – 100% identical outputs to the original model
* 📦 **>30% smaller model size** – lower VRAM requirements than the original model
* ⚡ **Compatible with ComfyUI** – drop-in support with custom nodes
* 🔧 **GPU-accelerated inference** – optimized for CUDA 12.1+
---
## Installation
### Requirements
* [ComfyUI](https://github.com/comfyanonymous/ComfyUI) installed
* NVIDIA GPU with **CUDA 12.1+**
### Setup
1. Install dependencies:
```bash
pip install -r requirements.txt
```
2. Install the DFloat11 custom nodes in ComfyUI:
```bash
cd /custom_nodes
git clone https://github.com/LeanModels/ComfyUI-DFloat11.git
```
---
## Usage
1. Once installed, the DFloat11 nodes show up under the `DFloat11` folder in the *Node Library*.
2. Download a DFloat11 model for ComfyUI from [Hugging Face](https://huggingface.co/DFloat11) and place it under `/models/diffusion_models`.
3. Drag and drop a `*.png` or `*.json` file from [workflows](https://github.com/LeanModels/ComfyUI-DFloat11/tree/master/workflows) into ComfyUI to load the workflow.
4. (Optional) Use the `DFloat11 Model Loader` node to load the model in `*.safetensors` format, which acts as a drop-in replacement for the `Load Diffusion Model` node.
---
## Resources
* 📖 [DFloat11 Paper (arXiv)](https://arxiv.org/abs/2504.11651)
* 🤗 [DFloat11 Models on Hugging Face](https://huggingface.co/DFloat11)
---
## Contributing
Contributions are welcome!
* Open an issue to request new model support
* Submit pull requests for bug fixes or improvements
[](https://github.com/LeanModels/ComfyUI-DFloat11/graphs/contributors)