Search papers, labs, and topics across Lattice.
This paper introduces Stream-R1, a reward distillation framework for streaming video generation that adaptively reweights the distillation objective based on both the reliability of student rollouts and the perplexity of spatiotemporal elements. By using a pretrained video reward model to rescale rollout losses and extract per-pixel gradient saliency, Stream-R1 concentrates optimization on reliable rollouts and challenging regions/frames. Experiments on standard benchmarks demonstrate consistent improvements in visual quality, motion quality, and text alignment compared to existing distillation methods.
Stop wasting compute on unreliable rollouts and easy frames: Stream-R1 adaptively focuses video diffusion distillation where it matters most, boosting quality without architectural changes or added inference cost.
Distillation-based acceleration has become foundational for making autoregressive streaming video diffusion models practical, with distribution matching distillation (DMD) as the de facto choice. Existing methods, however, train the student to match the teacher's output indiscriminately, treating every rollout, frame, and pixel as equally reliable supervision. We argue that this caps distilled quality, since it overlooks two complementary axes of variance in DMD supervision: Inter-Reliability across student rollouts whose supervision varies in reliability, and Intra-Perplexity across spatial regions and temporal frames that contribute unequally to where quality can still be improved. The objective thus conflates two questions under a uniform weight: whether to learn from each rollout, and where to concentrate optimization within it. To address this, we propose Stream-R1, a Reliability-Perplexity Aware Reward Distillation framework that adaptively reweights the distillation objective at both rollout and spatiotemporal-element levels through a single shared reward-guided mechanism. At the Inter-Reliability level, Stream-R1 rescales each rollout's loss by an exponential of a pretrained video reward score, so that rollouts with reliable supervision dominate optimization. At the Intra-Perplexity level, it back-propagates the same reward model to extract per-pixel gradient saliency, which is factored into spatial and temporal weights that concentrate optimization pressure on regions and frames where refinement yields the largest expected gain. An adaptive balancing mechanism prevents any single quality axis from dominating across visual quality, motion quality, and text alignment. Stream-R1 attains consistent improvements on all three dimensions over distillation baselines on standard streaming video generation benchmarks, without architectural modification or additional inference cost.
