Tripo P1: Image to Model
Game-ready low-poly from a single image
- image
- model_file
- model task_id
- GLB
Most text-to-3D and image-to-3D models give you a beautiful high-poly sculpture you then have to decimate into something a game will load. Tripo P1: Image to Model skips that whole ceremony. It's Tripo's "game-ready" model - a separate engine from the classic Tripo generation nodes, tuned specifically for low-poly, stable-topology meshes you can drop straight into an engine or a rigging pipeline.
Same Partner Node deal as the rest of the family: built into ComfyUI core, no install, runs on Tripo's cloud. Your image uploads, a job runs on their servers, and ComfyUI polls until the mesh is ready. You log in with a ComfyUI Account and pay from prepaid credits; the node's price badge estimates the run, and P1's textured modes cost more than its bare-geometry mode, so the estimate moves when you switch.
The inputs that matter
- image - the one required input. This node is about that source image, and it does best with a clean, well-lit, single-subject shot.
- output_mode - the key decision, and it's a dynamic combo that changes what else appears on the node. Geometry only returns an untextured mesh (cheaper, fastest). Textured adds color/PBR maps and unlocks extra widgets: a
pbrtoggle,texture_quality(standard/detailed),texture_alignment(fidelity to the source image vs alignment to mesh geometry), and anorientationchoice. - enable_image_autofix - pre-processes the input image for better generation quality. Worth turning on for rough or uncentered reference shots.
Then the advanced section: face_limit (target face count, 48–20,000; -1 lets Tripo pick adaptively - and adaptive is a sane default), auto_size (scale the output to approximate real-world meters - the tooltip for game-scale assets), export_uv (UV unwrap during generation; you can turn it off for faster geometry-only runs), and compress_geometry (geometry-based compression - and note the warning that you'll need to decompress before editing).
What comes out
The usual Tripo trio: a model_file string, a model task_id, and the GLB. Wire the GLB into Preview 3D & Animation to check it, or Save 3D Model to write it out. The task_id is the more interesting output - it chains into Tripo: Rig, Retarget, or Convert, so a single reference image can become a rigged, animated, engine-ready character without ever touching a DCC.
The catch to know about
P1 is priced per-call and the credits scale with your output_mode and texture_quality picks - geometry-only is the budget path, detailed texturing is the spendy one. The other thing to internalize is that face_limit is doing real work here: set it to your engine's budget instead of leaving everything to chance if you know what you're building for. Otherwise, this is the node I'd grab when the assignment is "props that must survive contact with a game engine."
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| output_mode | COMBO | "Geometry only" returns an untextured mesh. "Textured" adds color/PBR maps. | |
| enable_image_autofixopt | BOOLEAN | false | Pre-process the input image for better generation quality. |
| face_limitopt | INT | -1-1–20000 | Target face count, 48-20000. -1 lets Tripo pick adaptively. |
| model_seedopt | INT | 42 | — |
| auto_sizeopt | BOOLEAN | false | Scale the output to approximate real-world meters. |
| export_uvopt | BOOLEAN | true | UV unwrap during generation. Turn off for faster geometry-only runs. |
| compress_geometryopt | BOOLEAN | false | Apply geometry-based compression. Decompress before editing. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model_file | STRING | — |
| model task_id | MODEL_TASK_ID | — |
| GLB | FILE_3D_GLB | — |