Hy-Nomas generates multi-turn interactive video worlds from text and camera
Original title: Hy-Nomas: Towards Free-form Interactive World Generation
The 8B diffusion model uses action-aware causal attention and a prompt enhancer; authors release weights, code, and benchmarks.
Researchers present Hy-Nomas, a video world model with an 8B diffusion backbone that follows free-form text instructions and continuous camera control across multiple turns. It generates variable-length segments from current inputs and cached history, keeping descriptive context available while limiting action instructions to their time windows. A vision-language Prompt Enhancer turns requests and interaction history into conditions for the next segment. Training uses a multi-stage curriculum and diffusion distillation. The authors say human evaluation favors Hy-Nomas and release weights, code, benchmarks, and a lightweight adaptation workflow.
Unconfirmed: Claims of superior performance come from the authors' human evaluation.
- An 8B diffusion backbone generates variable-length segments from cached history.
- Descriptive captions stay available; action instructions are limited to intended windows.
- Curriculum adds causal generation, text, camera control, then model-generated histories.
- Weights, code, benchmarks, and an adaptation workflow are released.
From the original article
1 min read at huggingface.co
Interactive world generation requires models to follow successive commands while preserving relevant history and action consequences. We present Hy-Nomas, a video world model with an 8B diffusion backbone that supports free-form text instructions and continuous camera control across multiple interaction turns. It generates variable-length segments conditioned on current inputs and cached history.
By Qiaoling Chen, Jiaxiang Cheng, Xueqing Deng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.