#2727th of 52 in Open modelsFor image, video and design, 3d
trellis
Image- and text-to-3D asset generation model from Microsoft Research
Live demo ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 14.2k
- License
- MIT
- Last commit
- Oct 2026
- Language
- Python
12,048 on 10 Apr 202614,164 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
TRELLIS generates 3D assets from an image or text prompt using a structured latent (SLat) representation and rectified flow transformers. One latent decodes to 3D Gaussians, radiance fields or meshes, with GLB and PLY export. Pretrained models range from 342M to 2.0B parameters, and the repo includes training code and TRELLIS-500K dataset tooling.
Who it is for: Researchers and developers generating 3D assets on NVIDIA GPUs
Strengths
- One latent decodes to Gaussians, radiance fields or meshes; GLB and PLY export
- Training code, configs and TRELLIS-500K dataset tooling are included
- Supports variant generation, local editing and multi-image conditioning
- MIT license, weights on Hugging Face, hosted live demo
Weaknesses
- Requires an NVIDIA GPU with at least 16GB memory
- Tested only on Linux; Windows setup is not fully tested
- Install compiles several CUDA submodules and needs a CUDA Toolkit
- Text-conditioned models are less detailed than the image-conditioned model
What it needs
- GPU required
- Needs CUDA Toolkit 11.8 or 12.2, Conda, PyTorch 2.4.0, flash-attn or xformers, spconv, kaolin, nvdiffrast, Gradio (demo)
- Models: TRELLIS-image-large (1.2B), TRELLIS-text-base (342M), TRELLIS-text-large (1.1B), TRELLIS-text-xlarge (2.0B)
Quick answers
Is trellis open source?
Yes. Its code is on GitHub under the MIT license.
Does trellis need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 52| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 95 out of 100 |
| 2 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 87 out of 100 |
| 3 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 86 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 77 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 75 out of 100 |