fooocus

Fooocus is an offline, open-source image generation application built on the Stable Diffusion XL architecture and presented through a Gradio web interface. Its stated design goal is to let users focus on prompts and images rather than manual parameter tweaking, similar to hosted services such as Midjourney. The project describes an offline GPT-2 based prompt processing engine and sampling improvements intended to produce good results from short or long prompts. Key capabilities described in the README include text-to-image generation, upscaling and variation of input images, inpainting and outpainting (pan up/down/left/right), image prompts, face swap via InsightFace, image description, style presets, negative prompts, aspect ratio selection, and prompt weighting using A1111-style reweighting. It also documents its own inpaint algorithm and inpaint model, plus a native refiner swap inside a single k-sampler, negative ADM guidance, and a variation of self-attention guidance. Installation is offered for Windows via a downloadable package and run.bat, for Linux through Anaconda, Python venv, or native Python 3.10, for AMD GPUs via ROCm or DirectML (described as beta), for Mac via PyTorch MPS, and via Docker. A Colab notebook is also provided. The README states a minimum requirement of 4GB Nvidia VRAM and 8GB system RAM, with system swap required, and notes that CPU-only operation needs 32GB RAM. Three presets are available: default/general, realistic, and anime, each with different default models and configs. The UI can be exposed locally with --listen or shared via --share, and basic authentication can be added through an auth.json file. The project is described as being in limited long-term support with bug fixes only, with no current plans to migrate to newer model architectures. The README warns that many fake websites exist and that this repository is the only official source.