Head Forcing: Long Autoregressive Video Generation via Head Heterogeneity
Jiahao Tian, Yiwei Wang, Gang Yu, Chi Zhang † \dagger · 2026
Read Head Forcing: Long Autoregressive Video Generation via Head Heterogeneity by Jiahao Tian, Yiwei Wang, Gang Yu, Chi Zhang † \dagger (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-14487.txt.
Description
Tian and colleagues, writing in 2026, ask how autoregressive video diffusion can run longer without drifting. Their answer, Head Forcing, builds on head heterogeneity, treating attention heads as unequal workers, not one uniform pool. Local heads handle detail, anchor heads hold steady framing, and memory heads carry distant recall. Each group caches past tokens its own way. Local and anchor groups keep only vital tokens, while memory groups use a layered, refreshing store. The authors report clips stretching from 5 seconds to minute-level length, training-free, and topping rival methods.
Catalog record
| CREATOR | Jiahao Tian, Yiwei Wang, Gang Yu, Chi Zhang † \dagger |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75537 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.14487