Tencent Hy-Nomas: an open interactive video world model, now with ComfyUI nodes
Hy-Nomas is Tencent's open-source video world model: it builds a scene one video segment at a time, and you steer each new segment with text and camera moves while it remembers what came before. Community nodes published the same day run it natively in ComfyUI on 12 GB graphics cards, starting from your own image.
What it does
Each segment can add objects, change the style or scene, describe an action or move the camera, and a cache carries the world from one segment to the next. The distilled model uses 8 denoising steps for 720×1280 video at 24 fps, and Tencent ships a single RTX 4090 setup with INT8 weights, an INT8 cache, an FP8 decoder and CPU offload.
Results and limits
The ComfyUI-Hy-Nomas sample clips are 20 seconds at 1280×704, made of four chained 5-second segments that took about 3 minutes each on an RTX 4090. In drive mode, where you steer with the WASD keys, each one-second turn at 832×480 took about 4.3 seconds.
Worth knowing
The model itself has no image input; the ComfyUI nodes write your picture into the model's memory as the first segment so its look carries over. The nodes are marked work in progress and need ComfyUI v0.42 or newer, the weights are Apache 2.0, and Tencent runs a web demo on the project site.