PixelArtistry Trellis2 ComfyUI Installation Guide & Common Issues

The easiest way to install Trellis2 in ComfyUI is currently to use the ComfyUI-Easy-Install package because it will guarantee us that you will have all the right Python dependencies installed and it includes installation scripts for you, so you don't need to enter any commands yourself into a terminal line.

I have listed down below a step by step installation guide and common errors that people in our audience encountered and how you can fix them.

Before you start please make sure to read the requirements carefully:

• the installation currently only works with Nvidia RTX cards

• the installation currently only works on Windows (please check for Linux/Mac here)

• your GPU needs atleast 8gb of vram

Here is a step by step guide on how to install Trellis2 using the ComfyUI-Easy-Install package:

The scripts will install you the ComfyUI-Trellis2 repository by visualbruno: https://github.com/visualbruno/ComfyUI-Trellis2

1. Update your NVIDIA drivers.

2. Install ComfyUI-Easy-Install in a new folder https://github.com/Tavris1/ComfyUI-Easy-Install

3. Switch to Torch 2.8.0 by navigating to Add-Ons\Torch-Pack\Torch 2.8.0+cu128.bat.

4. Run the FlashAttention.bat file in the ComfyUI-Easy-Install\Addons folder,

5. Run the Trellis2 (requires Torch 2.8.0+cu128).bat in the ComfyUI-Easy-Install\Addons folder (It will download all files for you). This will download & install you then pytorch 2.8.0 with cuda 12.8

6. Since the latest ComfyUI version we need to add the --disable-dynamic-vram to our ComfyUI startup file. To do that go to ComfyUI-Easy-Install/Add-Ons/Tools and execute Toggle-DynamicVram.bat (make sure that the tag is enabled then)

7. Launch ComfyUI then by running the Start ComfyUI FlashAttention.bat

(In case you have 8gb vram add the --disable-pinned-memory to your comfyui startup .bat file.)

The workflows can be always found under each Trellis2 video in the video description. Alternatively you can also try the general workflow created by visualbruno by going to ComfyUI/custom_nodes/ComfyUI-Trellis2/example_workflows

When updating ComfyUI, please use first the Update Easy-Install.bat and then the Update ComfyUI.bat file in the ComfyUI-Easy-Install folder

Common installation issues

Repo id must use alphanumeric chars, '-', '_' or '.'. The name cannot start or end with '-' or '.' and the maximum length is 96:

Reason: not all files were correctly downloaded by the auto-downloader that is implemented in Trellis2 when launching a Trellis2

Fix:

Run this Checkpoint Downloader.bat file here in your ComfyUI folder which will automatically download all files for you or alternatively go to https://huggingface.co/microsoft/TRELLIS.2-4B/tree/main and download and copy all the files from this website and the ckpts folder to ComfyUI-Easy-Install/ComfyUI/models/Microsoft/TRELLIS.2-4B

CUDA error: an illegal memory access was encountered:

Fix:

since the latest ComfyUI version we need to add the --disable-dynamic-vram to our ComfyUI startup file. Alternatively go to ComfyUI-Easy-Install/Add-Ons/Tools and execute Toggle-DynamicVram.bat (make sure that the tag is enabled then).

Relaunch ComfyUI then afterwards

!!! Exception during processing !!! CUDA error: no kernel image is available for execution on the device:

Fix:

same as Cuda error above

HeaderTooSmall:

make sure that all models are correctly downloaded in ComfyUI-Easy-Install/ComfyUI/models/Microsoft/TRELLIS.2-4B

Prompt execution failed:

Fix:

1. Navigate to the Trellis2Load Model node

2. Where you see the TRELLIS.2-4B click the drop down and change it to the microsoft/TRELLIS.2-4B one

3. Launch the workflow and it will download the model automatically. Just make sure to not close comfyui then until it finished

Alert: Error loading model

Fix: You can ignore this message, it doesn’t mean your installation is broken. Simply close it.

Trellis2PreProcessImage:

index3 is out of bounds for axis with size 3

Fix:

• enable the “remove_background” option in the node itself or load an image with transparent background and RGBA channel