TRELLIS.2 by Comfy-Org

TRELLIS.2 by Comfy-Org

Comfy-Org's repackaged model files of Microsoft's TRELLIS.2-4B for ComfyUI, used in an image to model workflow, for people who build 3D assets from pictures.

Made by Comfy-Org 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.2 from Comfy-Org is a set of model files for ComfyUI, repackaged from microsoft/TRELLIS.2-4B. The files cover the main diffusion model, a vision encoder and two VAEs, one for shape and one for texture.

The repository links a ComfyUI workflow named Pixal3D and TRELLIS.2, Image to Model. That name points to making a 3D model from a picture, which is how the card frames it.

Who it suits: It suits ComfyUI users who want to try image-based 3D generation with files already arranged for the app. It is not a standalone program.

Industry | Who uses it | How they use it | Example
Games and media | 3D artists | Start a 3D asset from a reference image | A prop drafted from a concept picture
Product design | Designers | Turn a product photo into a first 3D model | A rough model of a gadget
Creative tools | ComfyUI users | Add 3D generation to their node workflows | An image to model graph
Research | Graphics researchers | Run the TRELLIS.2 model inside ComfyUI | Comparing the bf16 and int8 files

Skip it if you do not use ComfyUI, since these files are packaged for that app.

Where to use it:

• ComfyUI, in its models folders
• The Pixal3D and TRELLIS.2 image to model workflow that the card links
• Windows, macOS or Linux, with Python for the download step

Use case | Who | What they get back
Image to 3D model | 3D artists and designers | A 3D model made from an input picture
Workflow reuse | ComfyUI users | A ready graph to load and run

How to set it up: You need a few things in place first.

• ComfyUI installed on your computer
• Python and pip, for the download command
• Disk space for the files you choose

1. Install the Hugging Face hub package with pip.
2. Download the repository with the hf command.
3. Copy each file into the matching ComfyUI models folder, as listed below.
4. Load the image to model workflow in ComfyUI.

pip install -U huggingface_hub
hf download Comfy-Org/TRELLIS.2

The card lists where each file goes. All three folders sit inside the models folder of ComfyUI.

• The clip_vision folder holds dino_v3_vit_l.safetensors
• The diffusion_models folder holds trellis_2_int8_convrot.safetensors and trellis_2_bf16.safetensors
• The vae folder holds trellis_2_shape_vae_bf16.safetensors and trellis_2_texture_vae_bf16.safetensors

How to use it: You open ComfyUI, load the image to model workflow, point its nodes at these files and feed it a picture. The card gives no commands beyond the download, so the workflow itself guides the rest.

Tool | What it does
dino_v3_vit_l | The vision encoder that reads the input image
trellis_2_bf16 | The diffusion model in bf16 precision
trellis_2_int8_convrot | A diffusion model file with int8 in its name
trellis_2_shape_vae_bf16 | The VAE for shape
trellis_2_texture_vae_bf16 | The VAE for texture

Example: You place the five files in their ComfyUI folders and load the Pixal3D and TRELLIS.2 image to model workflow. You supply a picture of an object. The workflow runs and returns a 3D model of it.

For newcomers and experts: Start with the bf16 diffusion file and the linked workflow, and use one clear picture of a single object. Try the int8 file later if you want something smaller.

The repackaging splits the model into the pieces ComfyUI expects, so a vision encoder, a diffusion model and two VAEs load as separate nodes. The shape and texture VAEs suggest that geometry and surface are handled apart. The card gives no sizes or benchmarks, so test it on your own images.

Download the files with hf
Copy them into the ComfyUI folders
Load the image to model workflow
Give it a picture and get a 3D model

Good to know:

• License: MIT
• Cost to run: none named
• Data it reaches: it downloads files from Hugging Face onto your computer

On Hugging Face, 13 November 2026

• Library: diffusion-single-file
• Downloads in the last 30 days: 377,945
• Likes: 110
• License: mit
• Based on: microsoft/TRELLIS.2-4B
• Tags: diffusion-single-file, comfyui

Links

• Model: https://huggingface.co/Comfy-Org/TRELLIS.2
• Files: https://huggingface.co/Comfy-Org/TRELLIS.2/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Comfy-Org/TRELLIS.2

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