Face Swap Pro
Video face swap using inswapper_128, GFPGAN 1.4 enhancement at 512px, XSeg occlusion masking, BiSeNet face parsing, LAB colour matching and a feathered blend.
Consent and licences: only use photos and videos of people who have agreed to it. inswapper_128, buffalo_l and GFPGAN 1.4 are for non-commercial research use as far as I know, so check each licence before any other use. Label shared output as synthetic.
Hardware: CPU Spaces work but are very slow. Pick a GPU (for example T4) in Settings, and use short clips. On CPU hardware, change onnxruntime-gpu[cuda,cudnn] to onnxruntime in requirements.txt.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support