How to run OmniPart without a GPU
How can I run OmniPart without a GPU or a local install?
OmniPart has open weights, so you can run it yourself, but you need the hardware: a Linux NVIDIA setup like TRELLIS's, since no VRAM figure is published. Without a GPU, the simplest route is a hosted version. In OTOY Studio, OmniPart runs in the browser from 18 credits (€0.18) per run, with nothing to install, on the same canvas as 100+ other image, video and 3D models.
A part-aware image-to-3D model (SIGGRAPH Asia 2025). It generates a 3D object already split into separate, editable parts, and builds on TRELLIS.
Running it yourself
| Open weights | OmniPart is on Hugging Face. |
| License | MIT. Commercial use is allowed. |
| Hardware | The README does not state a VRAM requirement. Because it builds on TRELLIS, plan for a Linux NVIDIA setup like TRELLIS's. |
| Operating system | Linux with an NVIDIA GPU is the documented setup (Python 3.10). |
| ComfyUI | No. |
Sources: OmniPart on GitHub, OmniPart on Hugging Face. Checked October 16, 2026.
Run it yourself if: You need to change how parts are split (it takes a mask input locally), or run large batches.
Running it without a GPU in OTOY Studio
| Node | Takes | Credits per run |
|---|---|---|
| OmniPart | Prompt plus image | 18 (from €0.18) |
- Sign up free at https://app.otoy.studio. The free plan includes 100 credits a month, no credit card.
- Add a OmniPart node to a canvas and connect your inputs.
- Run it. It runs on OTOY Studio's hosted GPUs, so your machine only needs a browser.
- Chain the output into other models in the same workflow, or download it.
1 credit = €0.01. Prices are list prices before plan discounts (paid plans take 10% to 25% off every run). Model page: OmniPart.
Related
Hardware and license details are from each model's official repository and model card, checked October 16, 2026. Prices are computed from OTOY Studio's billing configuration, exported 2026-10-20.