SoL-Refiner: Speed-of-Light One-Step Refinement for High-Resolution Video

2026-09-29

AI Summary

This study addresses the prohibitive computational overhead of high-resolution video generation and the subsampling bottleneck inherent in conventional multi-step refinement approaches by proposing a one-step video refinement framework. The method introduces a three-stage training paradigm: high-resolution continual pre-training to enhance model capacity, reinforcement learning-based post-training to optimize generation quality, and one-step distillation to overcome inference speed limitations. Additionally, a dedicated evaluation benchmark, Refiner-Bench, is constructed. Experimental results demonstrate that the proposed approach achieves high-quality, single-step upscaling from low-resolution inputs to 4K videos. It outperforms existing external refiners on metrics such as VBench while accelerating inference latency by 8.91×, thereby significantly improving both the efficiency and fidelity of 4K video generation.

Abstract

High-resolution video generation is expensive, as its cost grows rapidly with the number of spatiotemporal tokens. A practical alternative first generates a lower-resolution video and then applies a refiner, but conventional multi-step refinement introduces a second sampling bottleneck. We present SoL-Refiner, a one-step video refiner that transforms low-resolution model outputs into 4K videos with a single denoising step. Our three-stage recipe combines high-resolution continual training, reinforcement learning (RL) post-training, and a final one-step distillation. We introduce Refiner-Bench, a video refinement benchmark constructed from the outputs of different video generators, and use a shared-input protocol to compare refiners at approximately 2K output resolution. At 2K, the one-step SoL-Refiner outperforms all external refiners on the VBench and UniPercept averages, while at $3840\!\times\!2176$ it improves both metrics over the three-step LTX-2.3 Refiner. With the complete acceleration stack, SoL-Refiner achieves an $8.91\times$ speedup in refinement latency over the same baseline in our 2K latency setting.

Problem

• high-resolution video generation
• video refinement
• sampling bottleneck
• computational cost

Innovation

• One-Step Video Refinement
• High-Resolution Video Generation
• Reinforcement Learning Post-Training
• Step Distillation
• Refiner-Bench