Disen-Forcing disentangles semantic anchoring from motion in video diffusion
Zhengyang Yu · hf · 2026-11-16
Researchers from ANU and Sony AI propose Disen-Forcing, a distillation-time method for autoregressive video diffusion models that disentangles attention-head specialization between semantic anchoring and motion, improving long-range appearance consistency while preserving motion dynamics over 30+ second generations.
More from Multimodal
- Dreamina's AI-generated cereal commercial trailer is indistinguishable from real TV ads — gen_ericai · 2026-11-16
- Keystone Attention node cuts H3 video-chain brightness loss from -28% to -9%, halves VRAM — roychodraws · 2026-11-16
- Seedance 2.5 video brings 1991 Nagasaki back to life, goes viral — SimplyAnnisa · 2026-11-16
- PixVerse demo: two mermaids having a drink in a sunken city tavern — Wise-Accident-9083 · 2026-11-16
- Blind test game suggests AI has passed the vocals Turing test too — chewygummie · 2026-11-16
- WeituAI's streaming video brain scores 69.9 on OVBench, beating Gemini baseline by 11 points — 机器之心 · 2026-11-16