Overview
Fooocus is an open-source, offline image generation application centered on usability: the project aims to reduce the need for manual parameter tuning so that users can “focus on prompting and generating.” It ships with a Gradio-based local web UI and automates model downloads and many setup steps to make the first-generation experience quick and simple.
Key features
- Built on Stable Diffusion XL (SDXL) architecture.
- Local/offline operation with a Gradio UI for running on desktop or remote servers.
- GPT-2–based prompt expansion (called “Fooocus V2”) to automatically expand or preprocess prompts for better quality with short prompts.
- Advanced sampling and a native refiner swap mechanism that reuses sampler momentum to improve high-resolution coherence.
- Image-focused workflows: img2img, upscaling (1.5x, 2x), subtle/strong variations, inpaint/outpaint with Fooocus-specific inpaint models, image prompt extraction, and face-swap features via InsightFace.
- Inline prompt features: wildcards, array processing (for generating multiple variants), inline LoRAs, prompt reweighting compatible with A1111-style syntax, and a negative prompt UI.
- Multiple presets (default, anime, realistic) and launchers (run.bat, run_anime.bat, run_realistic.bat) to simplify common use-cases.
- Low friction installation: prebuilt Windows package, Colab integration, Conda/venv instructions for Linux and macOS guidance (including MPS support for Apple silicon).
System & usage notes
- Project README documents minimal hardware requirements and platform-specific instructions. Typical minimum: 4GB GPU VRAM (Nvidia) and 8GB system RAM (with swap) for many presets; CPU-only mode is supported but much slower.
- Models and some optional components (e.g., Fooocus inpaint control model) are downloaded automatically on first run. The project provides direct release downloads for Windows builds.
- The README warns about many fake third-party sites impersonating Fooocus and explicitly states the GitHub repository as the only official source.
Project status & community
- The repository indicates a Limited Long-Term Support (LTS) stance focusing on bug fixes; no immediate plans to migrate to newer model architectures are announced in the README, though community forks and alternative UIs are referenced for experimenting with newer methods (e.g., WebUI Forge, ComfyUI forks).
- The project credits several contributors and downstream forks; it also thanks community members for models, styles, and tooling integrations.
Advanced and technical notes
- Fooocus implements several technical tricks tuned for SDXL, including negative ADM guidance, sampling adjustments, and adaptions designed to reduce overly-smooth outputs common in some XL runs.
- It supports numerous CLI flags for device configuration, FP precision, offloading, and network/listening options for multi-user or remote access.
Who is it for
Fooocus targets users who want high-quality, locally-run text-to-image and image-editing generation without heavy manual tuning. It is suitable for hobbyists and creators with modest GPUs who prefer offline, open-source workflows and for those who benefit from prompt-expansion and convenient presets.
Limitations & cautions
- The README highlights that Fooocus focuses on high-quality generation and thus keeps relatively high minimum requirements; it is not optimized for very small models to reduce resource needs.
- The project maintainers caution about fake external websites and recommend obtaining Fooocus only from the official GitHub repository.
