SVFR (Stable Video Face Restoration) is a unified framework for general video face restoration. It combines video face restoration (BFR), colorization, and restoration tasks by leveraging the generative and motion priors of Stable Video Diffusion (SVD) alongside task-specific information within a unified face restoration framework. This effectively integrates the complementary advantages of these tasks, enhancing temporal coherence and achieving superior restoration quality. The framework introduces learnable task embeddings to improve task recognition and adopts novel Unified Latent Regularization (ULR) to foster shared feature representation learning across different sub-tasks. Additionally, facial prior learning and self-referential refinement are introduced as auxiliary strategies to further enhance restoration quality and temporal stability. SVFR has achieved state-of-the-art results in video face restoration and sets a new paradigm for general video face restoration.