Fooocus

Fooocus is a minimalist Stable Diffusion XL desktop client released by lllyasviel in August 2023, built on Gradio. Its goal is to bring the Midjourney-style experience — "great images without elaborate prompts" — to a single local machine. With 50k+ stars on GitHub, it is widely recognized as the lowest-friction way to run SDXL.

The core revolves around two pillars: a low barrier to entry and strong output quality. First, an offline GPT-2 prompt processor turns short inputs like "house in garden" into richly detailed images, removing the keyword-stacking required by A1111-era workflows. Second, it maps Midjourney concepts to UI buttons — V1/V2/V3/V4, U1/U2/U3/U4, Inpaint/Up/Down/Left/Right, Image Prompt, --style, --stylize, --niji, --quality, --repeat — so migrating users face near-zero learning cost. Third, built-in Inpaint, Outpaint, and Image Prompt modules use the author's custom-trained SDXL inpaint and IP-Adapter models, avoiding the realism artifacts common in generic pipelines. Fourth, SDXL models are downloaded automatically: on first launch the chosen preset pulls the right checkpoints and inpaint models from Hugging Face.

Use cases include newcomers who want a download-and-run experience, designers who want Midjourney-quality output without subscribing to a cloud service, illustrators who need quick visual references, and second creators who need reliable SDXL inpainting on realistic faces. The onboarding bar is minimal: Windows users download the 7z archive, extract, and double-click run.bat. NVIDIA cards with 4GB of VRAM are enough to run it; the project also ships run_anime.bat and run_realistic.bat preset launchers.

Worth noting: the author states in the README that Fooocus is in Limited LTS — bug fixes only, no new model architectures. Users who need Flux and other newer models should switch to WebUI Forge or ComfyUI. Licensed under GPL-3.0, which is slightly more permissive than A1111's AGPL-3.0.