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.

Fooocus is designed to get out of the way of installation. On Windows, the entire process is a single download, a single extraction, and a double-click — no Python environment setup, no manual model downloads, and no configuration files to edit before your first generation. Linux, macOS, and Docker paths require a few more steps, but each platform follows a straightforward script-based workflow that handles model downloads automatically on first launch.

Minimum Hardware Requirements

The table below lists the minimum specifications required to run Fooocus locally on each supported platform. Devices below these specs may not be able to run the software.
Operating SystemGPUMin. GPU MemoryMin. System RAMSystem SwapNotes
Windows / LinuxNvidia RTX 4XXX4 GB8 GBRequiredFastest
Windows / LinuxNvidia RTX 3XXX4 GB8 GBRequiredUsually faster than RTX 2XXX
Windows / LinuxNvidia RTX 2XXX4 GB8 GBRequiredUsually faster than GTX 1XXX
Windows / LinuxNvidia GTX 1XXX8 GB (6 GB uncertain)8 GBRequiredOnly marginally faster than CPU
Windows / LinuxNvidia GTX 9XX8 GB8 GBRequiredFaster or slower than CPU
Windows / LinuxNvidia GTX < 9XXNot supported
WindowsAMD GPU8 GB8 GBRequiredVia DirectML; ~3× slower than Nvidia RTX 3XXX
LinuxAMD GPU8 GB8 GBRequiredVia ROCm; ~1.5× slower than Nvidia RTX 3XXX
MacM1 / M2 MPSSharedSharedShared~9× slower than Nvidia RTX 3XXX
Windows / Linux / MacCPU only0 GB32 GBRequired~17× slower than Nvidia RTX 3XXX
AMD GPU support on Windows uses DirectML. ROCm support on Windows is currently on hold. AMD support on all platforms is considered beta and is not intensively tested.

Platform Installation Guides

Windows

One-click .7z download with embedded Python. No existing Python installation required.

Linux

Anaconda, Python venv, or native system Python. AMD ROCm instructions included.

macOS

Runs via PyTorch MPS on Apple M1/M2 silicon with macOS Catalina or newer.

Docker

Containerised deployment. See the project’s docker.md for full compose configuration.

Default Models

Fooocus ships three preset configurations, each using a different checkpoint optimised for a different output style. The correct model is selected automatically based on which launcher or --preset flag you use.
PresetWindows LauncherLinux / CLI FlagMain ModelRefiner
General (default)run.bat(none)juggernautXL_v8Rundiffusion.safetensorsNot used
Realisticrun_realistic.bat--preset realisticrealisticStockPhoto_v20.safetensorsNot used
Animerun_anime.bat--preset animeanimaPencilXL_v500.safetensorsNot used
Model downloads are automatic. When you launch Fooocus for the first time with a given preset, the required checkpoint is downloaded directly to Fooocus\models\checkpoints. If you already have the files, copy them to that folder before launching to skip the download. If inpainting is used for the first time, an additional 1.28 GB inpaint patch (inpaint_v26.fooocus.patch) is downloaded automatically to Fooocus\models\inpaint.

Switching Presets After Launch

Starting with Fooocus 2.3.0, you can switch between presets directly in the browser UI without restarting the application. Two optional flags control this behavior:
# Disable the in-browser preset selector entirely
python entry_with_update.py --disable-preset-selection

# Automatically download missing models when switching presets
# (default behavior falls back to previously downloaded models)
python entry_with_update.py --always-download-new-model

Corrupted Model Files

If you see errors like MetadataIncompleteBuffer or PytorchStreamReader at launch, your model files are corrupted. Delete the affected .safetensors file and relaunch Fooocus to trigger a fresh automatic download.

Build docs developers (and LLMs) love