AppRisingofficial publisher

TRELLIS.2 (Microsoft)

App : A product you use directly

Microsoft image-to-3D model producing a PBR mesh in seconds, running locally.

The best freely commercialisable local 3D model (MIT).

Official site GitHub · 14,101 ★

Overall score

8.1 /10

Security: nothing to flag

Utility

9.0

Reliability

9.0

Security

No caveat on record

Ease

Technical · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Creating a quality 3D object from a single photo or a text prompt normally requires specialised software, hours of work and modelling skills. TRELLIS solves this by automatically generating a complete 3D asset in seconds, directly on your machine.

The model relies on an internal representation called SLAT (Structured LATent): a sparse 3D grid enriched with visual features extracted from multiple views. From an image or a text prompt, TRELLIS simultaneously produces a textured mesh (GLB), 3D Gaussians (PLY) and a radiance field, all exportable. The image-to-3D model has 1.2 billion parameters; text variants range from 342 million to 2 billion parameters.

TRELLIS targets developers, technical artists and researchers who want to integrate 3D generation into their pipelines without relying on a cloud service. The MIT licence allows free commercial use.

However, the tool is not suited for beginners: installation requires Linux (Windows partially supported), an NVIDIA GPU with at least 16 GB of VRAM, and familiarity with a Conda environment. The model excels at artistic-style assets; its ability to reproduce photorealistic real-world objects remains limited.

What it offers

·Image-to-3D generation: converts a photo into a textured mesh, 3D Gaussians and radiance field in a single pass.
·Text-to-3D generation: three models available (342M, 1.1B and 2B parameters) to create a 3D asset from a text description.
·Multi-format export: results can be exported as GLB (textured mesh), PLY (Gaussians) and MP4 preview videos.
·Local editing: targeted modification of a specific region of an existing 3D asset guided by a text or image prompt.
·Variant generation: production of multiple coherent versions of the same 3D object guided by text instructions.
·Multi-image conditioning: combining several input images to guide generation, without retraining the model.
·Gradio web interface: local demo accessible via browser, with no additional setup beyond the initial installation.

Our tips

Prefer the image-to-3D model over the text-to-3D model: results are more detailed and creative, as text training data is more limited.
For a single generation run, set the environment variable SPCONV_ALGO=native to skip the benchmarking phase at startup, which otherwise adds unnecessary delay.
On GPUs that do not support flash-attn (e.g. NVIDIA V100), install xformers only and set ATTN_BACKEND=xformers before running the model.
When exporting a GLB file, adjust the simplify parameter (ratio of triangles to remove) and texture_size according to the desired level of detail and your downstream pipeline constraints.
For text-to-3D generation, first generate an image with a text-to-image model, then use TRELLIS in image-to-3D mode to obtain better results.

Watch out

Windows installation is not officially supported and remains only partially tested through community contributions; issues may arise.
The model is limited to artistic-style assets; generating photorealistic real-world objects yields less convincing results.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: linux, windows
Host: comfyui
Licence: MIT
Runs locally: Yes, 16 GB VRAM recommended
FR interface: No

Security and data

Rating: Security: nothing to flag
Your data: Data stays local
Maintenance: Active

Links and repository checked on 25 October 2026 · added on 1 December 2025

How to install TRELLIS.2 (Microsoft)

1. Check prerequisites: Linux, NVIDIA GPU with at least 16 GB of VRAM, CUDA Toolkit (11.8 or 12.2), Conda and Python 3.8 or higher. 2. Clone the repository with its submodules: git clone --recurse-submodules https://github.com/microsoft/TRELLIS.git && cd TRELLIS 3. Run the setup script to create the Conda environment and install all dependencies: . ./setup.sh --new-env --basic --xformers --flash-attn --diffoctreerast --spconv --mipgaussian --kaolin --nvdiffrast 4. If multiple CUDA versions are installed, set the correct PATH before installation (e.g. export PATH=/usr/local/cuda-11.8/bin:$PATH). 5. To launch the local web demo, install the additional Gradio dependencies (. ./setup.sh --demo) then run python app.py. 6. Models are downloaded automatically from Hugging Face on first launch; no API key is required.

. ./setup.sh --new-env --basic --xformers --flash-attn --diffoctreerast --spconv --mipgaussian --kaolin --nvdiffrast

In the same category

Whole category 3D & printing →