Fooocus
lllyasviel · Images
GitHub stars
53,256Stats updated
What is Fooocus?
Fooocus concentrates on prompts and reference images instead of requiring a node graph or extensive sampler setup. Local installation and model downloads are still required. Its SDXL-focused tools cover variations, upscaling, inpainting, and outpainting, with the project now in limited, bug-fix-only long-term support.
Fooocus is an offline, open-source image generator built on Stable Diffusion XL. It emphasizes prompting and image generation over manual tuning, with a simplified installation path.
Core capabilities
- Generate SDXL images with built-in prompt expansion and style presets
- Create variations and image-guided edits with inpainting and outpainting
- Upscale outputs and switch among default, realistic, and anime presets
Prompt-first generation
Fooocus uses an offline prompt-processing engine, sampling adjustments, and curated styles to reduce the amount of tuning required before generation. The default interface keeps common choices prominent while leaving guidance, sampling sharpness, model selection, and other advanced controls available when needed.
- SDXL
- prompt expansion
- styles
- presets
Focused image operations
Input Image tools group variation, enlargement, inpainting, outpainting, and image prompting around a source image. Presets provide general, realistic, and anime starting points. Hardware support and performance vary; the README describes AMD support as beta and the Mac setup as less intensively tested.
- inpainting
- outpainting
- upscaling
- image prompt
Limited long-term support
The canonical README places Fooocus in limited long-term support with bug fixes only. It remains centered on Stable Diffusion XL and does not currently plan to adopt newer model architectures, so prospective users can evaluate a stable feature set without assuming ongoing model-family expansion.
- LTS
- bug fixes
- SDXL
Where it fits
Use cases
- 01
Create an image from a short prompt
Start with a concise description, let Fooocus expand and process the prompt, then select a style, aspect ratio, and quality mode without configuring a node graph or a large sampler panel.
- 02
Revise part of an existing image
Load an image for subtle or strong variation, mask an area for inpainting, extend the canvas through outpainting, or provide an image prompt to preserve visual direction while generating a new result.
- 03
Run a focused local SDXL setup
Install the Windows package or Python environment, keep the interface and processing on local hardware, and use the bundled default, realistic, or anime configurations as repeatable starting points.