TripoSplat by VAST-AI

TripoSplat by VAST-AI

Turns a single 2D image into a variable number of 3D Gaussians for asset creation, AR and VR, games and simulation. A model from VAST-AI for 3D creators and developers.

Made by VAST-AI on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: TripoSplat turns one 2D image into a set of 3D Gaussians, with a variable number of them depending on the picture. It was developed by TripoAI and published by VAST-AI. The card presents it as a pipeline tool for asset creation, AR and VR, game development and simulation environments.

Who it suits: It suits 3D artists, game developers and simulation builders who want a 3D starting point from a single picture, and developers who work with Gaussian splats.

Industry | Who uses it | How they use it | Example
Games | Game developers | Build 3D assets from a single concept image | Convert a picture of a prop into Gaussians
AR and VR | XR developers | Create 3D content for immersive scenes | Place a generated object into a scene
Simulation | Simulation builders | Populate environments with 3D objects | Generate objects for a test scene

Anyone who needs a full mesh pipeline with no Gaussians can skip it, since the output here is Gaussians.

Where to use it:

• The official TripoSplat inference code
• ComfyUI, through a workflow template for image to Gaussian splat
• Windows, macOS or Linux, where the files download with Python

Use case | Who | What they get back
Image to 3D asset | 3D artists | A set of 3D Gaussians for the object
Game or XR prototyping | Developers | A 3D object made from one picture
Simulation content | Simulation builders | Gaussians to place into an environment

How to set it up:

• Python with pip, for the download
• Either the official inference code or ComfyUI
• Five checkpoint files placed in the folder layout below

1. Install the Hugging Face hub client.
2. Download the repository's files.
3. Put each file in the right folder, under ckpts for the official code or under models for ComfyUI.
4. Open the inference code or the ComfyUI workflow template.

pip install -U huggingface_hub
hf download VAST-AI/TripoSplat

File | Folder
birefnet.safetensors | background_removal
triposplat_fp16.safetensors | diffusion_models
dino_v3_vit_h.safetensors | clip_vision
flux2-vae.safetensors | vae
triposplat_vae_decoder_fp16.safetensors | vae

How to use it: Place the files as shown, then run the official inference code or load the ComfyUI template. Give it one 2D image and it produces 3D Gaussians. The card does not show a command, so follow the inference code's own readme.

Example: You give the ComfyUI template a photo of a single object. The workflow returns a set of 3D Gaussians for it, and how many Gaussians you get varies with the image.

For newcomers and experts: A Gaussian splat is a 3D scene made of many small soft blobs. Start by placing the five files, opening the ComfyUI template and trying one clear picture of one object.

The repository holds several checkpoints that work together, including a diffusion model in fp16, a VAE decoder in fp16, a background removal file and a vision encoder. Its title mentions learned density control, which fits the variable number of Gaussians per image.

One 2D image goes in
The checkpoint files process it
Learned density control sets the Gaussian count
3D Gaussians come out

Good to know:

• Output: a variable number of 3D Gaussians, not a mesh
• Files: five safetensors files, laid out differently for the official code and for ComfyUI
• License: MIT
• Cost to run: none named
• Data it reaches: the images you convert and the local checkpoint files, with no web or account access named
• Not for: nothing named in the card

On Hugging Face, 13 November 2026

• Downloads in the last 30 days: 0
• Likes: 149
• License: mit
• Quantized versions: 5
• Fine-tunes: 3

Links

• Model: https://huggingface.co/VAST-AI/TripoSplat
• Files: https://huggingface.co/VAST-AI/TripoSplat/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download VAST-AI/TripoSplat

Then follow the usage in its model card. Nothing in this kit runs by itself.