Step-by-step generation, step-by-step control
Tencent's AI team Hunyuan has released a new video world model called Hy-Nomas. Its distinguishing trait is that it builds video piece by piece rather than generating it end to end in a single pass. Users can issue new instructions at each step, making it possible to change direction while generation is still under way.
Prompts and camera movement
Hy-Nomas can be steered with written prompts as well as camera movements. That sets it apart from conventional text-to-video tools: the result is not fixed by a single prompt but shaped as an interactive process.
Open-source emphasis
Tencent announced the model as open source. The term "world model" is used for approaches that try to preserve the physical and spatial consistency of a scene; such models are being evaluated for video generation as well as simulation and agent training.
Why it matters
Interactive video generation promises more control than one-shot generation. Open-source distribution also opens the model to scrutiny and testing by researchers and developers across different scenarios.
