LPM: Industrial-Scale Generative Video Restoration

2026-07-15

AI Summary

This work addresses the challenge of video degradation in complex real-world user-generated content by proposing the first industrially deployed generative video restoration system. Built upon diffusion models, the system integrates large-scale data engineering, an enhanced network architecture, a progressive training strategy, and a temporal pyramid inference mechanism to achieve high-fidelity, temporally consistent restoration of videos of arbitrary length. The approach strikes a breakthrough balance between perceptual quality and inference efficiency. Deployed on Kuaishou, it now covers approximately 45% of total watch time, significantly improving key user experience metrics; at equivalent visual quality, it reduces bitrate by 20%, yielding annual bandwidth cost savings of hundreds of millions of RMB, and has been successfully integrated into products such as Kling.

Abstract

We present the Large Processing Model (LPM), a diffusion-based generative framework for photorealistic video restoration under complex, in-the-wild degradations. To our knowledge, LPM is the first generative video restoration model deployed at industrial scale. LPM addresses the diverse degradations in user-generated content (UGC) through a unified system encompassing large-scale data engineering, foundation-model training, and efficient inference. Its enhanced architecture, progressive training strategy, and temporal-pyramid inference mechanism jointly enable high-fidelity, temporally consistent restoration of arbitrarily long videos across the broad content distribution encountered on UGC platforms. LPM has been deployed in production at Kuaishou, where videos processed by the model account for approximately 45% of total viewing time, delivering consistent improvements across key quality-of-experience metrics. Beyond perceptual enhancement, LPM delivers substantial system-level benefits: at comparable perceptual quality, it reduces bitrate by 20% relative to Kuaishou's in-house codec, yielding annual bandwidth cost savings on the order of hundreds of millions. Its low serving cost also enables integration into products such as Kling, demonstrating that generative restoration can be practical, scalable, and cost-effective for large-scale video processing.

Problem

• video restoration
• user-generated content
• in-the-wild degradations
• generative model
• industrial-scale

Innovation

• generative video restoration
• diffusion model
• temporal-pyramid inference
• industrial-scale deployment
• bitrate reduction