41 upvotes
Hy-Nomas lets you steer video worlds with text and camera
What is Hy-Nomas: Towards Free-form Interactive World Generation?
Hy-Nomas is an 8-billion-parameter diffusion model that can generate video environments on the fly. Users can give free-form text prompts and move the camera continuously, and the model keeps track of past actions to maintain consistency.
Who it is for
Game developers and interactive-simulation creators can prototype dynamic scenes without hand-crafting assets.
How to try it
Open the demo on the HuggingFace Spaces page linked from the paper.