Fooocus is a free, open-source, offline image generation application built on Gradio and powered entirely by the Stable Diffusion XL (SDXL) architecture. It is designed around a single idea: you should only have to think about your prompt and your image — not about samplers, CFG scales, or model weights. Installation is deliberately minimal, with fewer than three mouse clicks separating a fresh download from your first generated image, and a minimum GPU memory requirement of just 4 GB (Nvidia).Documentation Index
Fetch the complete documentation index at: https://mintlify.com/lllyasviel/Fooocus/llms.txt
Use this file to discover all available pages before exploring further.
What Makes Fooocus Different
Rather than exposing every knob SDXL offers, Fooocus bakes a carefully tuned set of improvements directly into the generation pipeline so that results are consistently high quality without manual intervention:- Offline GPT-2 prompt expansion — An on-device GPT-2 model automatically enriches your prompt before sampling, similar to Midjourney’s hidden pre-processing. This means even a short prompt like
"house in garden"produces polished, detailed imagery. - Custom inpaint algorithm and models — Fooocus ships its own inpaint control model (
inpaint_v26.fooocus.patch), producing results that are consistently more coherent than software using the standard SDXL inpaint method. - Custom IP-Adapter (Image Prompt) — The Image Prompt feature uses a proprietary algorithm that improves both result quality and prompt understanding compared to standard SDXL IP-Adapters or Revision methods.
- Negative ADM guidance — Because XL’s highest resolution level lacks cross-attention, positive and negative signals can be too similar, leading to plastic or overly smooth outputs. Fooocus compensates by modifying the ADM conditioning on both sides.
- Self-Attention Guidance (SAG) — A carefully tuned, very low-weight variation of SAG is applied as a final safeguard against overly smooth textures, using an anisotropic Gaussian kernel for better structure preservation.
- Native refiner swap inside one k-sampler — When a refiner model is used, Fooocus swaps from base to refiner within a single k-diffusion sampling loop, reusing accumulated momentum for more coherent final images.
- A1111-compatible prompt weighting — Prompts copied directly from Civitai work as expected using Automatic1111’s reweighting algorithm, including embedding syntax like
(embedding:file_name:1.1).
Feature Comparison: Fooocus vs. Midjourney
The table below maps Midjourney features to their Fooocus equivalents, drawn directly from the official documentation.| Midjourney | Fooocus |
|---|---|
| High-quality text-to-image without needing much prompt engineering or parameter tuning. (Unknown method) | High-quality text-to-image without needing much prompt engineering or parameter tuning. (Offline GPT-2 prompt processing engine and sampling improvements — results are always beautiful, whether your prompt is "house in garden" or 1,000 words.) |
| V1 V2 V3 V4 | Input Image → Upscale or Variation → Vary (Subtle) / Vary (Strong) |
| U1 U2 U3 U4 | Input Image → Upscale or Variation → Upscale (1.5x) / Upscale (2x) |
| Inpaint / Up / Down / Left / Right (Pan) | Input Image → Inpaint or Outpaint → Inpaint / Up / Down / Left / Right (Fooocus’s own inpaint algorithm and models) |
| Image Prompt | Input Image → Image Prompt (Fooocus’s own image prompt algorithm) |
--style | Advanced → Style |
--stylize | Advanced → Advanced → Guidance |
--niji | Multiple launchers: run.bat, run_anime.bat, run_realistic.bat. Supports SDXL models from Civitai. |
--quality | Advanced → Quality |
--repeat | Advanced → Image Number |
Multi Prompts (::) | Use multiple lines of prompts |
| Prompt Weights | I am (happy:1.5) — uses A1111’s reweighting algorithm |
--no | Advanced → Negative Prompt |
--ar | Advanced → Aspect Ratios |
| InsightFace | Input Image → Image Prompt → Advanced → FaceSwap |
| Describe | Input Image → Describe |
Project Status
Fooocus is currently in Limited Long-Term Support (LTS) mode. Future updates will focus exclusively on bug fixes. There are no current plans to migrate to or incorporate newer model architectures such as Flux. The existing feature set is considered functionally complete and stable. For workflows requiring newer architectures, consider WebUI Forge or ComfyUI.
Get Started
Quickstart
Generate your first image on Windows in under 5 minutes — no configuration required.
Installation
Full installation instructions for Windows, Linux, macOS, and Docker.