What it does

This repo is a localized fork of the well-known FaceSwap project, offering a Python-based pipeline to extract faces from images or video, train a deep-learning model on them, and swap one face onto another. It runs on Windows, Linux, and macOS, though the README makes clear you’ll want a modern CUDA-capable GPU; AMD support is partial at best.

The interesting bit

The README spends nearly as much space on an ethics declaration as on technical instructions—a notable choice for a project in a space notorious for misuse. The authors frame the tool as an educational entry point into AI, explicitly condemning non-consensual or deceptive use. Whether that stance travels with the code is, of course, up to whoever runs it.

Key highlights

• Three-stage pipeline: extract faces, train a model, convert source media
• Optional GUI via python faceswap.py gui
• Built-in video conversion helper (tools.py effmpeg) or manual ffmpeg workflow
• Reusing pre-trained models significantly speeds up results
• Suggests starting with lookalikes if your training data is thin

Caveats

• The repo appears to be a translation/mirror rather than original development; the Travis badge still points to the upstream deepfakes/faceswap project
• “Partial” AMD GPU support likely means frustration for non-NVIDIA owners
• The colloquial sign-off (“你懂得，小屌丝！”) suggests this isn’t an official or particularly maintained fork

Verdict

Worth a look if you read Chinese and want a localized, pre-packaged introduction to deepfake tooling. Skip it if you’re after novel research or guaranteed AMD compatibility.