Before any wardrobe, pose, body, or style expansion, capture the subject's face from every reference angle. These close-ups become the identity anchor the AI reuses across front, side, back, clothing layers, medical overlays, lighting modes, and cinematic renders — so the same person shows up in every downstream frame.
The base identity string reused across every FUSI reference shot.
Every parametric avatar must export the largest native-resolution face reference available before any wardrobe, pose, body, or style expansion runs against it.
Export at the largest available native-resolution image before any downstream generation. Never use low-resolution screenshots when a direct export path exists.
Use lossless PNG whenever possible. Avoid JPEG re-compression — chroma subsampling and block artifacts destroy the fine facial detail the identity anchor depends on.
The face should occupy the majority of the frame so the model can read eye spacing, brow ridge, nose bridge, nostril shape, cheekbones, lips, jawline, chin, ears, hairline, skin tone, and natural asymmetry.
Do NOT crop the chin, forehead, ears, or hairline in a primary reference. Cropping is only allowed for explicitly secondary detail shots (e.g. the ear/jawline close-up).
Solid outer border = frame edge · dashed gold = keep-in-frame zone. Never crop into the dashed line.