Hy-Nomas: A free-form interactive video world generation model
In brief
- Hy-Nomas generates interactive videos with free-form text commands and continuous camera control.
- The model uses causal attention to separate descriptive context from action instructions.
- Code, weights, and benchmarks are published for reproducible research.
Researchers introduce Hy-Nomas, an interactive video generation model featuring an 8-billion-parameter diffusion backbone. According to the announcement, it supports free-form text instructions and continuous camera control across multiple interaction turns. The system generates variable-length video segments based on current inputs and cached history. Its architecture relies on causal attention aligning structured video captions with actions, where descriptive context remains globally accessible while action instructions are confined to their intended temporal windows. A vision-language Prompt Enhancer (PE) converts user requests and multimodal interaction histories into structured conditions for the next segment. The authors outline a progressive training curriculum including causal generation, text interaction, camera control, and adaptation to model-generated histories, followed by diffusion distillation.
In human evaluation, Hy-Nomas outperforms existing interactive world generation models. Model weights, code, benchmarks, and a lightweight adaptation workflow are released to support reproducible research in this field.
This summary gives you the essentials. The full article, its figures and quotes are on Hugging Face Daily Papers.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error