Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation

TL;DR

This paper introduces Reward Forcing, a novel framework for streaming video generation that improves motion dynamics and long-term consistency using EMA-Sink tokens and Rewarded Distribution Matching Distillation, achieving state-of-the-art results.

Contribution

The paper proposes EMA-Sink for better long-term context and introduces Re-DMD to prioritize dynamic content in distillation, enhancing motion quality in streaming video generation.

Findings

Achieves state-of-the-art performance on benchmarks.

Generates high-quality streaming videos at 23.1 FPS.

Effectively maintains long-term consistency and dynamic motion.

Abstract

Efficient streaming video generation is critical for simulating interactive and dynamic worlds. Existing methods distill few-step video diffusion models with sliding window attention, using initial frames as sink tokens to maintain attention performance and reduce error accumulation. However, video frames become overly dependent on these static tokens, resulting in copied initial frames and diminished motion dynamics. To address this, we introduce Reward Forcing, a novel framework with two key designs. First, we propose EMA-Sink, which maintains fixed-size tokens initialized from initial frames and continuously updated by fusing evicted tokens via exponential moving average as they exit the sliding window. Without additional computation cost, EMA-Sink tokens capture both long-term context and recent dynamics, preventing initial frame copying while maintaining long-horizon consistency. Second, to better distill motion dynamics from teacher models, we propose a novel Rewarded Distribution Matching Distillation (Re-DMD). Vanilla distribution matching treats every training sample equally, limiting the model's ability to prioritize dynamic content. Instead, Re-DMD biases the model's output distribution toward high-reward regions by prioritizing samples with greater dynamics rated by a vision-language model. Re-DMD significantly enhances motion quality while preserving data fidelity. We include both quantitative and qualitative experiments to show that Reward Forcing achieves state-of-the-art performance on standard benchmarks while enabling high-quality streaming video generation at 23.1 FPS on a single H100 GPU.