Running Fooocus Via Containerization In Linux

[Hamza-b93]

Hi.

Recently, I set about trying to set-up Fooocus in a containerized environment on Linux. I have managed to set this environment up and run it locally on my pc. For context, these are the specifications of my computer:

OS: Arch Linux
Kernel: x86_64 Linux 6.12.10-arch1-1
CPU: AMD Ryzen 7 3700X 8-Core @ 16x 4.426GHz
GPU: AMD Radeon RX 6600 (radeonsi, navi23, LLVM 19.1.7, DRM 3.59, 6.12.10-arch1-1)
GPU Drivers Installed: vulkan-radeon, vulkan-amdgpu-pro, xf86-video-amdgpu

Dockerfile Setup:
dockerfile.txt

Building The Container

podman build -t fooocus .

Starting The Container:

podman run -it --rm \ -v fooocus-models:/app/models \ --device=/dev/kfd \ --device=/dev/dri \ -p 7865:7865 \ fooocus

Downloading Models

1. Option A: Let Fooocus Handle It:

After starting Fooocus, it will automatically download the required models to /app/models/checkpoints. If you mapped the fooocus-models volume, these models will persist across container runs.

2. Option B: Manual Download:

Download the default models from the Fooocus repository and place them in the mounted volume:
podman volume inspect fooocus-models

Copy the downloaded models into the models/checkpoints directory inside the volume.

Access Fooocus:

http://localhost:7865

Optional: Launch Preset Modes:

- Anime Edition:

podman run -it --rm \ -v fooocus-models:/app/models \ --device=/dev/kfd \ --device=/dev/dri \ -p 7865:7865 \ fooocus --preset anime

- Realistic Edition:

podman run -it --rm \ -v fooocus-models:/app/models \ --device=/dev/kfd \ --device=/dev/dri \ -p 7865:7865 \ fooocus --preset realistic

Notes And Troubleshooting:

1. Verify GPU Access: Ensure that /dev/kfd and /dev/dri are accessible and that your user is in the video group:
sudo usermod -aG video $USER
newgrp video

2. Model Volume: Use the fooocus-models volume to store models persistently across container runs.

3. AMD GPU Beta Support: Fooocus for AMD GPUs is in beta.

That is mostly it! This basic containerization template should get you going on Linux with AMD GPUs.
Suggestions and improvements are always welcome!