Stable Diffusion WebUI Forge is a community-driven fork of the popular Stable Diffusion WebUI, focused on optimizing resource management, accelerating inference, and integrating experimental features. It targets users seeking enhanced performance and advanced capabilities beyond the original WebUI, offering a more dynamic development platform.

How It Works

Forge is built upon SD-WebUI 1.10.1, incorporating a custom "Flux" backend that leverages advanced techniques like BitsandBytes NF4 and GGUF quantization for efficient model loading and inference. It features a Gradio 4 UI and supports features like asynchronous swapping of model components between CPU and GPU, enabling larger models to run on less VRAM.

Quick Start & Requirements

• Install: Download and run the provided one-click installation package (includes Git and Python).
• Prerequisites: CUDA 12.1+ and PyTorch 2.3.1 are recommended. Alternative CUDA/Torch versions are available.
• Setup: Uncompress the package, run update.bat for the latest version, then run.bat.
• Docs: Flux Tutorial, Flux Tutorial 2

Highlighted Details

• Natively supports various quantized model formats (NF4, GGUF Q8_0/Q5_0/Q4_0) with GPU weight sliders and async swapping.
• Integrates FreeU V2 for improved image quality and detail.
• Features a Gradio 4 UI with enhanced canvas functionality and Wacom pressure support.
• Offers improved LoRA handling, including skipping patching and single-load options.

Maintenance & Community

The project is actively maintained, with regular updates and a responsive developer addressing reported issues. Users can find community support and news via linked resources.

Licensing & Compatibility

The project appears to be under a permissive license, compatible with commercial use and closed-source linking, though specific license details should be verified.

Limitations & Caveats

Some features, such as Microsoft Surface touch pressure support for the canvas and ControlNet Union/Flux implementations, are noted as broken or not yet implemented. OFT LoRA support is also currently broken. The UI and functionality are under active development and subject to change.