A text-driven face video hair editing method based on timing compensation
By adopting a text-driven face video hair editing method based on time-compensation, the problems of temporal continuity, inaccurate text mapping, and facial attribute destruction in existing technologies are solved. This method achieves high-precision hair editing and facial integrity, supports editing of multiple hairstyles and hair colors, and eliminates inter-frame flickering.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HENAN NORMAL UNIV
- Filing Date
- 2026-04-18
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for text-driven editing of hair in real-life facial videos suffer from issues such as loss of temporal continuity, inaccurate text-visual mapping, easy destruction of facial attributes, and insufficient preservation of identity and background attributes, resulting in poor editing effects.
A text-driven face video hair editing method based on temporal compensation is adopted. By combining the inverse mapping coding module, CLIP cross-modal coding module, temporal compensation decoupling module, Hair Mapper module and generator module, the method achieves accurate mapping between text semantics and hair features and inter-frame coherence. The method also combines temporal compensation and feature fusion modules to handle occluded areas and avoids erroneous modification of non-hair areas.
It achieves high-precision matching between hair editing results and text semantics, improves inter-frame coherence, fully preserves facial and background attributes, has a high degree of automation, supports editing of multiple hairstyles and hair colors, has an error of less than 5% in matching editing results with text semantics, eliminates inter-frame flickering, and significantly improves identity similarity and background integrity.
Smart Images
Figure CN122420601A_ABST