Practical overview
When it fits
Use it when you want to work with a code-based research implementation for generating 3D Gaussian objects or avatars from prompts, rather than a hosted generation service.
Who it is for
- Researchers exploring text-to-3D generation and Gaussian-splat representations.
- 3D creators prototyping objects or avatars from text prompts.
- Developers preparing generated assets for a Unity workflow using UnityGaussianSplatting.
Documented uses
- Generate a 3D object from a text description.
- Create a 3D avatar from a text prompt.
- Explore how 2D and 3D diffusion can contribute to a Gaussian-splat scene.
- Prepare generated assets for import into Unity through a separate tool.
Requirements & limits
The project is publicly available on GitHub under the Apache-2.0 license. The documented generation workflow requires local setup and a GPU; no minimum GPU specification is given.
- The documented installation pins PyTorch 2.0.1+cu117 and related CUDA 11.7 packages, alongside Gaussian Splatting components and Shap-E.
- A fine-tuned Shap-E model must be downloaded separately and placed in ./load.
- Object and avatar generation use different prompt settings; avatar examples include
system.load_type=1. - Unity import is documented through UnityGaussianSplatting, a separate project.
- The README describes real-time rendering of generated instances, while its generation command runs as a training job.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up GaussianDreamer for prompt-based generation
Prepare the documented dependencies, model file and repository so the text-to-3D or text-to-avatar commands can be used.
Before you start
- Python environment capable of installing the documented packages
- PyTorch 2.0.1+cu117, torchvision 0.15.2+cu117 and torchaudio 2.0.2 from the CUDA 11.7 package index
- Ninja and the repository requirements
- 3D Gaussian Splatting rasterization and simple-knn components
- OpenAI Shap-E and the fine-tuned Shap-E model file placed in ./load
Steps
Install the specified PyTorch, torchvision and torchaudio packages, then install Ninja and the repository requirements.
Clone GaussianDreamer and install its diff-gaussian-rasterization and simple-knn submodules.
Clone Shap-E and install it in editable mode.
Download the fine-tuned Shap-E model and place it in the repository’s ./load directory.
Expected outcome
The README’s listed dependencies and model file are prepared for the documented launch commands.
Guide limitations
- The README does not specify a minimum GPU model or memory requirement.
- This recipe covers setup only; successful installation or generation is not independently verified.
Launch a text-to-3D object generation job
Use the repository’s documented command pattern to request a 3D object from a text prompt.
Before you start
- GaussianDreamer setup and fine-tuned Shap-E model file completed
- A GPU available to the launch command
Steps
From the GaussianDreamer repository, run the documented launch command with the
gaussiandreamer-sd.yamlconfiguration, training enabled, GPU 0, and a text prompt.Replace the example prompt with the object description to generate.
For the README’s Unity-import variant, add
system.sh_degree=3to the command.
Expected outcome
The command starts the documented text-to-3D workflow; the README describes generated assets as importable into Unity with UnityGaussianSplatting.
Guide limitations
- The README provides command examples but no guaranteed completion time for a particular run or prompt.
- Unity import uses the separate UnityGaussianSplatting project.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- hustvl/GaussianDreamer · Official homepage title · retained in its source language