VividFace: Face-Aware One-Step Diffusion for High-Fidelity Video Face Enhancement

Video Face Enhancement (VFE) aims to restore high-quality facial details from degraded video sequences, enabling a wide range of practical applications. Despite substantial progress in the field, current methods that primarily rely on video super-resolution and generative frameworks face three fundamental challenges: (1) computational inefficiency caused by iterative multi-step denoising in diffusion models; (2) insufficient recovery of fine-grained facial textures; and (3) poor restoration quality due to the lack of high-quality face video training data. To address these challenges, we propose VividFace, a one-step diffusion framework that reformulates a text-to-video generation model into a single-step generator with switchable face-aware guidance that persistently focuses optimization on perceptually critical facial regions across both latent and pixel spaces. Furthermore, we propose a human-aligned MLLM-driven data curation pipeline that leverages the video understanding capabilities of Multimodal Large Language Models (MLLMs) and iteratively refines scoring criteria through lightweight human-in-the-loop calibration to align quality judgments with human perceptual expectations, yielding a high-quality face video dataset MLLM-Face90. Extensive experiments demonstrate that VividFace achieves superior performance in perceptual quality, identity preservation, and temporal consistency across both synthetic and real-world benchmarks, while achieving a $12\times$ inference speedup over state-of-the-art diffusion-based VFE methods.