TRELLIS-image-large

TRELLIS-image-large

Generates a 3D asset from a single image, using Microsoft's large image-conditioned TRELLIS model. Made for developers, game artists and researchers who want 3D content from pictures.

Made by microsoft 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: TRELLIS-image-large takes a picture and generates 3D content from it. It is the version of Microsoft's large TRELLIS 3D generator that is steered by an image. A research paper on structured 3D latents introduced the model.

Who it suits: It suits developers, 3D artists and researchers who want to turn pictures into 3D assets without modeling each one by hand.

Industry | Who uses it | How they use it | Example
Game development | 3D artists | Start a 3D asset from a concept picture | Turning a prop sketch into a first 3D draft
E-commerce | Product teams | Make a 3D view from a product photo | A rotating view built from one product picture
Research | Graphics researchers | Study image-to-3D generation | Comparing results across input pictures
Design | Prototype makers | Mock up a physical object quickly | A rough 3D shape from a reference image

Skip it if you need a 3D model from text alone, since this version is conditioned on an image.

Where to use it:

• Python on Windows, macOS and Linux, to download the files
• The TRELLIS code release, which the card names as its code home
• Any workflow that needs a 3D asset made from a picture

Use case | Who | What they get back
Image to 3D asset | 3D artists | A generated 3D result from one picture
Rapid prototyping | Designers | A rough 3D shape of an object

How to set it up:

• Python, to run the download command
• The TRELLIS code, which the card points to for running the model
• A picture to convert

1. Install the Hugging Face tools.
2. Download the model files to your computer.
3. Follow the TRELLIS code release for how to run the model on a picture.

pip install -U huggingface_hub
hf download microsoft/TRELLIS-image-large

How to use it: You give the model one image and it generates a 3D result. The part of the card available here shows no run command, so use the TRELLIS code release and its project page for the exact steps.

Tool | What it does
TRELLIS-image-large | Generates 3D content from an input image
TRELLIS code release | Holds the code that runs the model

Example: You give the model a clear picture of a single object. It generates a 3D version of that object. The card does not name the output formats, so check the TRELLIS code release for what you can export.

For newcomers and experts: This model turns a flat picture into a 3D object. A good first try is to follow the TRELLIS code release with one clear, simple object picture.

If you know the field, the card describes TRELLIS as a large 3D generative model built on structured 3D latents, and this is its image-conditioned version. The card gives no sizes or speeds, so read the paper for how it works underneath.

Provide one picture
Model reads the image
Model generates 3D content
Export or inspect the result

Good to know:

• Scope: this version takes an image as input.
• Details: the card is short, and the paper has the technical detail.
• License: MIT.
• Cost to run: none named.
• Data it reaches: it downloads model files from Hugging Face and works on the pictures you give it.
• Not for: the card names no limits, so check the code release for hardware needs.

On Hugging Face, 13 November 2026

• Task: image to 3d
• Library: trellis
• Downloads in the last 30 days: 2,762,068
• Likes: 722
• License: mit
• Languages: en
• Quantized versions: 3
• Fine-tunes: 14
• Tags: trellis, image-to-3d

Links

• Model: https://huggingface.co/microsoft/TRELLIS-image-large
• Files: https://huggingface.co/microsoft/TRELLIS-image-large/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download microsoft/TRELLIS-image-large

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