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.

Windows is the easiest way to get started with Fooocus. The official release ships as a self-contained .7z archive that includes an embedded Python runtime — no Python installation, no virtual environment, and no dependency management required. Download the archive, extract it, and double-click a batch file to start generating images.

​
Hardware Requirements

The table below shows the minimum hardware needed to run Fooocus on Windows. System Swap (Virtual Memory) is required in all cases.
GPUMinimal GPU MemoryMinimal System MemoryNotes
Nvidia RTX 4XXX4 GB8 GBFastest
Nvidia RTX 3XXX4 GB8 GBUsually faster than RTX 2XXX
Nvidia RTX 2XXX4 GB8 GBUsually faster than GTX 1XXX
Nvidia GTX 1XXX8 GB (6 GB uncertain)8 GBOnly marginally faster than CPU
Nvidia GTX 9XX8 GB8 GBFaster or slower than CPU
Nvidia GTX < 9XXNot supported——
AMD GPU8 GB8 GBVia DirectML, ~3× slower than RTX 3XXX

​
Installation Steps

1

Download the release archive

Download the latest Fooocus Windows package:Fooocus_win64_2-5-0.7zAlways download from the official GitHub releases page. There are many fake Fooocus websites — the GitHub repository is the only legitimate source.
2

Extract the archive

Open the .7z file with 7-Zip and extract it to a folder of your choice (for example, C:\Fooocus). Do not extract to a path inside Program Files or any folder that requires elevated permissions.
3

Launch Fooocus

Inside the extracted folder, run one of the following batch files by double-clicking it:
Batch filePresetMain model
run.batGeneral (default)juggernautXL_v8Rundiffusion
run_anime.batAnimeanimaPencilXL_v500
run_realistic.batRealisticrealisticStockPhoto_v20
You can also switch presets directly in the browser after Fooocus 2.3.0 without restarting.
4

Wait for automatic model download

On the first launch, Fooocus will automatically download the required Stable Diffusion XL checkpoint to Fooocus\models\checkpoints. The download size is several gigabytes — allow time for it to complete before the web UI opens. If you already have the model files, copy them into that folder to skip the download.If you later use the Inpaint feature, Fooocus will also download inpaint_v26.fooocus.patch (~1.28 GB) to Fooocus\models\inpaint the first time you inpaint an image.Once the UI loads, open your browser to http://127.0.0.1:7865.

​
AMD GPU Support (DirectML)

AMD GPU support on Windows uses the torch-directml backend. It is in beta and approximately 3× slower than an Nvidia RTX 3XXX. ROCm for Windows is currently on hold.
AMD GPUs on Windows are supported via DirectML. After extracting the archive, open run.bat in a text editor and replace its contents with the following: 
.\python_embeded\python.exe -m pip uninstall torch torchvision torchaudio torchtext functorch xformers -y
.\python_embeded\python.exe -m pip install torch-directml
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --directml
pause
Then save the file and run it. For the Anime or Realistic presets on AMD, use the corresponding launch lines: 
.\python_embeded\python.exe Fooocus\entry_with_update.py --directml --preset anime

.\python_embeded\python.exe Fooocus\entry_with_update.py --directml --preset realistic

​
Troubleshooting

Nvidia Driver 531 performance issue — Nvidia drivers above version 532 have been reported to be up to 10× slower than driver 531 for some configurations. If generation is unexpectedly slow, try rolling back to driver 531: Laptop | Desktop.
System Swap / Virtual Memory — Fooocus requires Microsoft’s Virtual Swap (paging file) to be enabled. Windows enables this automatically in most cases. If you see a RuntimeError: CPUAllocator error, verify that Virtual Memory is turned on and that each drive has at least 40 GB of free space. You can find the Virtual Memory settings under System Properties → Advanced → Performance Settings → Advanced → Virtual Memory.
MetadataIncompleteBuffer / PytorchStreamReader errors — If you see either of these errors at startup, one or more model files are corrupted, most likely due to an interrupted download. Delete the affected files from Fooocus\models\checkpoints (or the relevant model subfolder) and relaunch Fooocus to trigger a fresh download.

Build docs developers (and LLMs) love

Get started for freeTalk to us