Bringing Native Support for 3D Gaussian Splats into ComfyUI with TripoSplat
Turn one image into 3D Gaussians with flexible detail levels for any scene.
We're excited to announce that TripoSplat, an open-source model from Tripo, is supported on day 0 in ComfyUI. Upload a single image and get a 3D Gaussian asset you can view and use in modern 3D pipelines.
TripoSplat is especially good at stylized subjects such as characters, props, and creative designs where look and detail matter. Unlike many 3D generators that produce the same amount of detail everywhere, TripoSplat leverages a novel approach for adaptive density control, to put more detail where your image needs it and stay lighter on simple areas, so you get optimal results without wasting file size or rendering cost.
Highlights
Open source (MIT): weights and inference code are available for local runs, customization, and community workflows.
One image in, 3D splat out: no multi-angle photo shoot required.
Smarter detail placement: richer geometry where it counts, simpler where it does not.
You choose the detail level: use fewer Gaussians for background props, more for hero assets, or export several versions for different devices (like level-of-detail in games).
Where It Fits
Modern 3D pipelines: 3D previews for stylized characters, environments, props, and creative designs.
3D-to-2D guidance: block out a 3D concept, product, or scene and render it to guide image or video generation.
AR/VR & interactive: turn a concept image into something you can explore in 3D.
Examples
Get Started in ComfyUI
Update ComfyUI to the latest version v0.23.0 (Cloud and Desktop will follow soon)
Open the Template Library and search for TripoSplat.
Choose the TripoSplat workflow template.
Follow the instructions in the template to download the models
Upload an image, then run the workflow
Model weights: 🤗 VAST-AI/TripoSplat
Inference code: VAST-AI-Research/TripoSplat
We look forward to seeing what you build with TripoSplat.
As always, enjoy creating!
is working on cloud now, or i need to wait?
Would be nice if you could still upload multiple image to get even better results.