// the find
deepfakes/faceswap
Deepfakes Software For All
A full desktop pipeline (extract faces -> train a model -> convert footage) for the original deepfake use case, with a Tkinter GUI and a dozen interchangeable model architectures (DFaker, IAE, Phaze-A, Villain, etc.). Built for hobbyists and VFX tinkerers who want to run face-swap training locally rather than call a hosted API.
The extract/train/convert stages are each proper plugin systems (plugins/extract, plugins/train) rather than one monolithic script, so swapping detectors, aligners, or model architectures doesn't mean forking the codebase. GPU backend support is genuinely broad - lib/gpu_stats has separate modules for nvidia, rocm, apple_silicon, and cpu fallback, which is more hardware coverage than most ML repos bother with. It has an actual pytest suite wired into CI and a readthedocs site with per-module API docs, not just a README. Still getting commits in August 2026 after being basically synonymous with the original 2017 deepfake tooling - most repos from that era are abandoned.
No PyPI package - setup is clone-the-repo-and-run-a-shell-script via INSTALL.md, and dependency pinning around CUDA/ROCm versions is the kind of thing that rots within a year. The GUI is Tkinter, which caps how good the actual user experience can be regardless of how much polish goes into lib/gui/theme.py. 'Modern GPU with CUDA' is a hard requirement for anything beyond toy results, so this isn't something you spin up on a laptop to evaluate. The README leads with an ethics manifesto defending against misuse, which signals real reputational/legal exposure for anyone building on top of it rather than just running it standalone.