DDRL anchors diffusion RL with data: 1M+ GPU hours show it beats reward hacking in video generation
gowthami_s · x · 2026-10-15
gowthamis highlights the arXiv paper "Data-regularized Reinforcement Learning for Diffusion Models at Scale" (Ye, Zou, Ermon, Liu et al.). DDRL anchors the RL policy to an off-policy data distribution via forward KL, combining reward maximization with standard diffusion loss to fight reward hacking (quality degradation, over-stylization, lost diversity). With 1M+ GPU hours and 10k+ double-blind human evals, it achieves the highest human preference on high-res video generation while baselines like DanceGRPO/FlowGRPO game the reward with unrealistic outputs.
Related event: DDRL keeps diffusion loss during RL to fix reward hacking in video models(5 posts)→
More from Multimodal
- FastVideo's FastH3-Omni 8-step distilled model trends on Hugging Face with ComfyUI support — FastVideo · 2026-10-23
- Kylon Launches Annotations to Analyze and Remix Any Video Into Your Own Brand Version — testingcatalog · 2026-10-23
- Droste-effect AI image with infinite nested scenes stuns: 'You can see everything?' — yungcontent · 2026-10-23
- AI-generated Slavic folklore horror anthology 'Midnight Ryazh' debuts episode 1 — Creative_Dependent_6 · 2026-10-23
- Japan's ai& launches video generation with MiniMax H3: 12-second clips from $0.0125/sec — DavidBennett__ · 2026-10-23
- Canva unveils Craft, its first in-house image model that plans before generating — The Verge AI · 2026-10-23