Abstract: Highlights•PHiD improves facial identity consistency in pose-guided human animation videos.•Pose-Driven Face Morphing uses 3DMM to generate proxy faces for temporal consistency.•Masked Face Adapter refines facial details via proxy embedding and masked attention.•Facial ID-Preserving Loss enhances identity retention using feature and pose cues.•PHiD generalizes well and boosts facial consistency in pose-guided video generation.
Loading