Video generation improves speed and quality with condition shifting
Few-Step Streaming Video Generation via Condition Shifting
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Generating videos step-by-step can be slow and require extra computing power. The authors found a way to speed up this process by using a technique called condition shifting, which cleverly reuses parts of the system instead of needing a separate network. This makes video creation faster and uses less memory, while still keeping the video quality and how well the video follows instructions high. People who watched the videos preferred the new method for following instructions and overall look, though some motion aspects favored the older system.
What this means in practice
- •For video content developers: Create longer and higher-quality instructional videos faster with reduced computational resources by integrating condition shifting techniques.
- •For interactive media engineers: Build streaming video systems that adaptively balance video quality and generation speed for real-time applications using condition shifting.
Authors
Deyuan Liu, Jungang Li, Kaiwen Zheng, Kai Jiang, Jintao Chen, Yaole Wang, Ming Lin, Tao Lin, Jianfei Chen, Jun Zhu, Jintao Zhang
Abstract
Few-step sampling reduces the cost of streaming video generation. Distribution Matching Distillation (DMD) supervises few-step generators with a separate fake-score network, which adds training memory and computation and must track the evolving generator. We extend APEX to streaming video by using condition shifting to estimate the velocity field of the generated distribution within the generator. The original text condition selects generation, while a shifted condition selects fake estimation using the same parameters. This removes the independent fake network while retaining S2-Avatar's frozen reference teacher, Self-Replay Forcing, and perceptual regularization. We evaluate APEX on 180-second videos using three backbone evaluations per block and a one-step spatial Refiner, with a 70-case human study and a 320-case StreamAV-Bench comparison. Compared with S2-Avatar, APEX achieves higher video instruction fulfillment on the Progressive track (3.856 versus 3.831) and higher visual aesthetics across both tracks (0.641898 versus 0.632606). Human raters favor APEX most strongly on instruction following, with +30.0% net preference, followed by consistency and video quality at +4.3% each. Motion quality favors S2-Avatar, audiovisual synchronization is predominantly tied, and usability is unchanged. Separate measurements under matched training settings show 4.4% lower peak allocated GPU memory per GPU and 9.2% shorter mean iteration time than DMD.