Face-swapping in a box, four ways

What it does This repository is Docker scaffolding for the FaceFusion platform—packaging, not the engine itself. It provides compose files to run the face-swapping and lip-syncing application (per its GitHub topics) in containers targeting CPU, NVIDIA CUDA, TensorRT, or AMD ROCm.

The interesting bit Rather than detecting hardware at runtime, the project splits backends into completely separate compose stacks, each pinned to its own localhost port. You choose your target—CPU on 7865, CUDA on 7870, TensorRT on 7875, or ROCm on 7880—and launch the matching variant.

Key highlights

• Discrete compose files for CPU, CUDA, TensorRT, and ROCm
• Static port allocation per backend to prevent collisions
• Docker Hub badges track FaceFusion 3.7.0 images
• Released under the OpenRAIL-S license

Caveats

• The README is a bare quickstart; substantive documentation lives entirely at docs.facefusion.io
• No application code lives here—just the container orchestration to run the upstream platform

Verdict Grab it if you need FaceFusion in a container and want backend-specific launch templates without runtime auto-detection. Skip it if you are hunting for the core application logic or in-repo documentation.