Skip to main content

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.

This guide covers the fastest path to a working image on Windows. Fooocus ships as a self-contained .7z archive that includes an embedded Python runtime — no existing Python installation, no virtual environments, and no manual package management required. If you are on Linux, macOS, or Docker, see the Installation page for platform-specific instructions.
You need at least 4 GB of Nvidia VRAM and 8 GB of system RAM to run Fooocus on Windows. Microsoft’s Virtual Swap is used to extend available memory and is enabled by default on most Windows installations.

​
Steps

1

Download the Windows release

Download the official Fooocus Windows package from GitHub:Fooocus_win64_2-5-0.7zThis archive contains a fully embedded Python environment — no separate Python installation is needed.
Only download Fooocus from the official GitHub release page linked above. Many fake websites impersonate Fooocus in search results. There is no official fooocus.com, fooocus.net, fooocus.ai, or similar domain.
2

Extract and launch

Extract the .7z archive using 7-Zip or a compatible tool. Once extracted, open the Fooocus folder and double-click run.bat.A terminal window will open and Fooocus will begin its startup sequence. After a moment, your default browser will open automatically to the Gradio UI at http://127.0.0.1:7865.
3

Wait for the automatic model download

On the very first launch, Fooocus downloads the default checkpoint automatically:
  • juggernautXL_v8Rundiffusion.safetensors is saved to Fooocus\models\checkpoints
Download time depends on your internet connection — the file is several gigabytes. You can watch progress in the terminal window. Subsequent launches skip this step entirely.
If you already have this model file from a previous install, copy it into Fooocus\models\checkpoints before running run.bat to skip the download entirely.
4

Type a prompt and click Generate

Once the UI is open in your browser:
  1. Type a description of the image you want in the prompt box — for example, a red fox sitting in a snowy forest at dusk.
  2. Leave all other settings at their defaults.
  3. Click the Generate button.
Fooocus will automatically expand your prompt using its offline GPT-2 engine, apply the default styles (Fooocus V2, Fooocus Enhance, Fooocus Sharp), and produce the image. Generated images are saved to the Fooocus\outputs folder.

​
Launcher Variants

After your first successful generation, you can explore the two additional preset launchers included in the same folder:
LauncherPresetModel Downloaded
run.batGeneral (default)juggernautXL_v8Rundiffusion.safetensors
run_anime.batAnimeanimaPencilXL_v500.safetensors
run_realistic.batRealisticrealisticStockPhoto_v20.safetensors
Each launcher automatically downloads its own required model on first use. You can also switch between presets without restarting by using the preset selector in the browser UI (available since Fooocus 2.3.0).

​
Performance Tips

Nvidia driver versions above 532 have been reported to be up to 10× slower than driver version 531 in some configurations. If generation times seem unusually long, consider rolling back to:
On a mid-range laptop with a 6 GB Nvidia RTX 3060 and 16 GB system RAM, Fooocus runs at approximately 1.35 seconds per iteration at the default Speed performance preset (30 steps). For faster generation at the cost of some quality, try switching to Extreme Speed (8 steps via LCM LoRA), Lightning (4 steps), or Hyper-SD (4 steps) in the Advanced → Quality section.

​
What’s Next

Windows Advanced Options

AMD GPU setup, Virtual Swap configuration, and custom model paths on Windows.

Installation

Full platform coverage — Linux, macOS, Docker, and hardware requirements.

Build docs developers (and LLMs) love

Get started for freeTalk to us