Disen-Forcing: Disentangling Semantic Anchoring from Motion for Autoregressive Video Diffusion A paper titled "Disen-Forcing: Disentangling Semantic Anchoring from Motion for Autoregressive Video Diffusion" proposes a method to stabilize long-horizon rollouts in autoregressive video diffusion models (AR-VDMs), which the paper says degrade rapidly once the inference horizon outruns the training horizon due to compounding exposure bias. The work frames existing frame-sink approaches as heuristic and separates semantic anchoring from motion to address the degradation. The paper's abstract excerpt does not disclose benchmark figures or quantitative results. Autoregressive video diffusion models AR-VDMs enable real-time, long-horizon video generation but degrade rapidly once the inference horizon outruns the training horizon, due to compounding exposure bias. To stabilize long rollouts, recent methods heuristically adopt frame sink, keeping initial fr