Achieving simultaneous preference alignment and distillation acceleration in video diffusion models remains an open challenge. Existing methods optimize the two objectives over mismatched representation spaces, where improving one objective often compromises the other. To overcome this, we propose Reward Lightning, a unified framework that aligns and accelerates a video diffusion model within a single shared representation. Its central principle is homology: both objectives are evaluated on identical latent features, which mitigates the gradient conflicts that arise when they are optimized over disjoint representations. As a foundational component, we first introduce a latent reward model (LRM) that scores videos directly in the latent space, without decoding back to the pixel space. Building on the LRM, homologous preference distillation (HPD) reuses this shared backbone to perform adversarial distillation and preference alignment jointly, yielding few-step generators that remain faithful and well aligned. Extensive experiments demonstrate that the LRM surpasses pixel-level and latent-level reward baselines by 11.0% and 14.7% in preference accuracy, and that Reward Lightning generates high-fidelity videos in merely 1 to 4 steps, improving the average VBench score by 2.1% while leading in text alignment, motion quality, and visual quality.
TL;DR
Reward Lightning is a unified framework that simultaneously achieves preference optimization and distillation acceleration for large-scale video diffusion models.
Its core insight is homogeneity that is, homogeneous structures and homogeneous data.
It can generate high-fidelity, human-aligned videos in 1–4 NFEs.
Motivation:
(a) Heterogeneous Training: Pixel rewards cause gradient conflicts that shift distillation trajectories.
(b) Disjoint Training: Sequential training leads to catastrophic forgetting of preference distributions.
(c) Homologous Training: Shared homologous structures and data guide the generator towards jointly optimal distributions.
