Fooocus Download: Simple, Prompt-Focused AI Image Generation

Fooocus, by lllyasviel (also the creator of ControlNet), was designed so you can type a prompt and get a good image without learning dozens of settings. Note that the project is in limited long-term support with bug fixes only: for newer model families such as Flux, the developers point to other tools.

What does Fooocus do?

• Easy prompts: Its built-in “Fooocus V2” prompt expansion, an offline GPT-2-based engine, improves short prompts automatically.
• Image tools: Image prompts, inpainting, outpainting, upscaling and variations.
• Prompt features: Prompt weights, multi-line prompts, wildcards and inline LoRAs.
• Presets: Default, anime and realistic editions, plus support for SDXL models from community sites.
• Offline: After the models download, everything runs on your computer.

Is Fooocus right for you?

Fooocus suits beginners who want attractive SDXL images with minimal effort. If you need the newest models, look at ComfyUI or SD.Next instead.

How to install

Windows: The README links a 7z package (Fooocus_win64_2-5-0.7z). Download it, extract it and run run.bat. The default models download automatically on first launch. Get it only from the official GitHub page; the project warns that sites such as fooocus.com, fooocus.net and fooocus.ai are fake.

Linux (Python 3.10):

git clone https://github.com/lllyasviel/Fooocus.git
cd Fooocus
python3 -m venv fooocus_env
source fooocus_env/bin/activate
pip install -r requirements_versions.txt
python entry_with_update.py

On Apple Silicon Macs, the project recommends installing conda and a PyTorch nightly build with MPS support first. Add --preset anime or --preset realistic to launch the other editions; on Windows, run_anime.bat and run_realistic.bat do the same.

Minimum hardware

• NVIDIA RTX 2000/3000/4000: At least 4 GB of VRAM and 8 GB of RAM (swap required).
• NVIDIA GTX 10xx: 8 GB of VRAM; only marginally faster than the CPU.
• AMD: 8 GB of VRAM, via DirectML on Windows or ROCm on Linux.
• Mac: M1/M2 via MPS, about 9 times slower than an RTX 3000-series card.
• CPU only: Possible with 32 GB of RAM, but about 17 times slower than an RTX 3000-series card.

Fooocus is free and open source under the GPL-3.0 license.