lllyasviel/fooocus

Fooocus is a free, offline, open-source image generation tool that lets you create high-quality images from text descriptions. It is designed to give users the experience of a polished commercial service like Midjourney, but running entirely on your own computer with no accounts, no subscriptions, and no internet connection required. The central design philosophy is simplicity: instead of exposing dozens of technical parameters (like those found in tools such as Stable Diffusion WebUI or ComfyUI), Fooocus hides most of the complexity and focuses the user on just writing a prompt, a text description of the image they want. Under the hood it uses a Stable Diffusion XL (SDXL) model, which is a large neural network trained on billions of images to generate realistic and artistic pictures from text. Fooocus adds its own pre-processing pipeline (based on GPT-2, a language model) to automatically improve prompts, along with custom sampling algorithms that reliably produce high-quality results even with short or simple prompts. Beyond basic text-to-image generation, it supports inpainting (editing specific parts of an existing image), outpainting (extending an image beyond its original borders), image-to-image variations (generating alternatives based on an uploaded photo), face swapping, and style presets similar to Midjourney's style flags. You would use Fooocus if you want a quick and beginner-friendly way to generate images locally, without deep technical configuration. It requires a minimum of 4 GB of GPU memory on an Nvidia card. The tool is currently in limited maintenance mode (bug fixes only), as the underlying SDXL architecture it is based on has been largely superseded by newer models. The tech stack is Python, with the web interface built using Gradio (a Python library for building interactive ML demos), and the image generation backend running PyTorch.