modellisted

# Image-to-3D — Hunyuan3D-2.1 → GLB Generate an untextured 3D mesh from a single image. The shape pipeline runs inside the ComfyUI container; files pass through ComfyUI's bind-mounted input/output dirs, so the GLB lands on the host automatically. ## Requirements Hunyuan3D-2.1 must be provisioned in the container. If it isn't, `gen_model.py` stops — run **`/setup:hunyuan`** once to install it (clone + deps; the shape weights auto-download on first use). ## Generate python3 "${CLAUDE_PLUGIN_ROOT}/scripts/gen_model.py" \ --image <path-to-image> [--name <basename>] \ [--steps 50] [--guidance 5.0] [--seed 1234] [--octree 256] [--decimate 0] - Background is removed automatically (rembg) before reconstruction. - `--decimate N` collapses the mesh to ~N faces and flat-shades it (needs Blender on the host); 0 keeps the dense mesh. - ~1 minute per mesh on the GPU. On `cudaErrorDevicesUnavailable`, restart the container: `docker restart "$FORGE_COMFY_CONTAINER"`. Prints `glb: <host-path>` (and `glb_decimated:` when decimating). When the user wants it off the box, use **`/forge:publish`** on the GLB — don't publish automatically. ## Typical flow `forge:image` → choose a result → `forge:model --image <that file>`.