Image → 3D model
A textured 3D model from one photograph.
What this is
One photograph becomes a textured GLB. It runs Hunyuan3D-2 through our own handler rather than a ComfyUI graph, so the file below is the handler's input object, not something you can open in ComfyUI.
Running it yourself
The worker is in the repo at gpu/hunyuan3d-worker/ — a Dockerfile that builds Hunyuan3D-2 with its texture rasteriser and the Blender Python module. Build it, push it, and create a RunPod serverless endpoint from the image with a 16 GB+ GPU. Sign up at RunPod if you need an account.
Pass --build-arg HY3D_BAKE=0 and point HF_HOME at a network volume if you would rather the weights lived there than in a 25 GB image.
Not switched on here yet — the image has to be built and pushed first.
You give it
- One picture of the subject
- Optionally a caption, and whether to cut the background out first
You get back
- A GLB you can spin in the browser, download, or attach to a tool
One object, photographed straight on, filling most of the frame. A cluttered picture becomes a cluttered mesh.