✦
Summary
ID-V2V is an identity-preserving video restylization framework from Eyeline Labs for applying scene, lighting, and style changes to a source video while keeping human identity and performance stable. It targets production settings where facial likeness, expression, gaze, and lip synchronization must survive heavy visual edits.
Key Features
Restylizes source videos while preserving facial identity and performance.
Uses edited keyframes to guide scene style, lighting, and background changes.
Supports relighting as a constrained version of the broader restylization task.
Handles imperfect keyframes by re-aligning motion after the first frame.
Uses facial normal maps and relit facial regions as preservation controls.
Supports single-subject and multi-subject video examples.
Provides project materials with paper, code link, and demo videos.
Targets production-quality video editing and visual storytelling research.
About ID-V2V
The method decouples edit-driven synthesis from source-grounded identity preservation. It uses relit facial regions, facial normal maps, edited keyframes, and depth sequences so the model can propagate a new style or lighting setup while re-aligning to the original performance after imperfect keyframes.
ID-V2V is useful for filmmakers, VFX teams, and researchers exploring controllable video-to-video generation. It supports restylization and relighting workflows where the creative look can change substantially without sacrificing the actor's facial motion and identity.