FADRA: Frequency-Aware Diffusion with Residual Adaptation for Video Face Restoration
FADRA recovers high-quality facial details from degraded videos using a frequency-aware diffusion framework with iterative residual adaptation. It adapts a pre-trained text-to-video model efficiently and uses a repeated residual head guided by low-quality latent information to refine details step-by-step. This approach significantly improves facial structure recovery and temporal consistency compared to existing methods.