WORKFLOW · #identity · low–mid vram

Face Swap — FaceFusion · HyperSwap

Studio-grade identity transfer: FaceFusion's hyperswap_1a_256 swapper at 1024px pixel-boost with CodeFormer enhance and occlusion masking — far beyond the classic inswapper-128.

Before / After

targetFaceFusion swap

Workflow Preview

Recipe — FaceFusion

facefusion.py headless-run -s source.jpg -t target.jpg -o out.png \
  --processors face_swapper face_enhancer \
  --face-swapper-model hyperswap_1a_256 --face-swapper-pixel-boost 1024x1024 \
  --face-enhancer-model codeformer --face-enhancer-blend 70 \
  --face-mask-types region occlusion --face-mask-blur 0.3 \
  --face-detector-model yolo_face --execution-providers cuda

FaceFusion is a standalone tool (its own venv), not a ComfyUI graph — this is the headless command.

⬇ ReActor workflow.json ⬇ API format

How It Works

Two paths to a face swap. The portfolio path (shown in the before/after) is FaceFusion + HyperSwap: the modern hyperswap_1a_256 swapper run at a 1024×1024 pixel-boost with CodeFormer enhancement and region+occlusion masking — sharp, well-blended, and a genuine identity change. The in-graph path is the ComfyUI ReActor node (download above) for when you want the swap inside a node graph. Both keep the target's hair, pose and lighting; only the face identity changes.