Your Data Manifold is Secretly a Reward Model: Shell-LCC for Text-to-Video Generation

Recent text-to-video (T2V) diffusion models rely heavily on auxiliary rewardsignals (e.g., via reward models or DPO) to align generated content with human aesthetics andimprove realism. These signals, however, incur substantial computational overhead, requirecostly human annotations, and often yield limited improvement in fine-grained local details. Inthis paper, we argue that your data manifold is secretly a reward model. By explicitlymodeling the manifold structure of high-quality Supervised Fine-Tuning (SFT) data andencouraging video latents to lie on this manifold, we derive dense, differentiable, and nearlycost-free reward signals that significantly improve video quality, particularly in mitigating low-level distortions. Our modeling builds upon Local Coordinate Coding (LCC), which capturesthe ‘skeleton’ of the manifold. However, directly applying LCC suffers from mean regression,pulling latents toward the geometric mean and losing high-frequency details. We thereforeextend it to Shell Local Coordinate Coding (Shell-LCC), which models the manifold ‘surface’as an isotropic shell to align with the true high-density region. Experiments demonstratethat our approach improves realism, enhances high-frequency details, reduces over-smoothingartifacts, and alleviates motion blur.