VisageLab

Realistic face-swapping is sensitive to alignment quality and blending artifacts far more than to runtime speed, so VisageLab's research focus went into getting the preprocessing and blending stages right rather than optimizing raw swap speed. It's a computer-vision research project exploring realistic face-swapping using the InsightFace model family, with a systematic visual QA step built in since face-swap realism is easy to overestimate from spot-checking alone.

Preprocessing normalizes and aligns faces across source images, the InsightFace-based pipeline handles the swap with blending refinement, and batch processing runs it efficiently across image sets. The result: a working research prototype demonstrating a practical, end-to-end face-swap pipeline from raw input to blended output.