New h3-vbvr model on HuggingFace tackles image-to-video

On HuggingFace, on August 18, 2026, the model Patarapoom/h3-vbvr was published with the image-to-video task. At the time of publication, it had 2,434 downloads and not a single "like". The code is released as open source.

What it means

The model relates to image generation but solves a different task than the tools in our AI for 3D section. All three generators we cover — OpenLRM, Hunyuan3D 2.1, and TRELLIS — work on the image-to-3d task: an image goes in, and a 3D model comes out. h3-vbvr, as described by the author, does image-to-video, meaning it generates video from an image rather than a three-dimensional object. There is no direct overlap with our fact sheets: none of the tools we cover solves the image-to-video task. Comparing h3-vbvr with them on characteristics like export formats or VRAM would be incorrect — these are different classes of tasks. The model appeared recently, and there is no data about its author, architecture, or limitations in the source. For those looking for tools to generate 3D models from images, h3-vbvr is not suitable — it solves an adjacent but different task.