Best modern face-swap workflow if I have 100+ high-quality face photos?
Im trying to get better face swaps than FaceFusion/ReActor alone. FaceFusion works, but results often look fake/plastic and not fully natural.
I have 100+ high-quality photos of my own face: front, side angles, cropped face, full body, different lighting, expressions, etc.
My idea is:
- Train a face LoRA from my dataset
- Use ReActor/FaceFusion only for the basic swap/alignment
- Then use FLUX/ComfyUI inpaint + LoRA to refine only the masked face
- Goal: keep target pose/body/hair/background/lighting, but make the face more realistic and closer to my identity
Is this the best current approach, or is there a better 2026 workflow?
Also, should I use:
- ReActor blended face model + LoRA refine?
- IPAdapter/FaceID/PuLID?
- FLUX Fill / Qwen Image Edit?
- pure LoRA inpaint?
- some other ComfyUI workflow?
Main goal: realistic photo face swap, not full image generation.