OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Worldgen

WorldGen is an open-source Python project that generates explorable 3D scenes from text prompts or images, with documented options to save Gaussian splats or export meshes.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Worldgen — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use WorldGen when you want a locally run, code-oriented workflow for generating and exploring scenes from prompts or images, or exporting generated results as PLY assets.

Who it is for

  • Developers prototyping text-to-scene or image-to-scene generation.
  • Technical artists and researchers preparing 3D environments for games, simulations, robotics or virtual reality.

Documented uses

  • Generate an indoor or outdoor 3D scene from a text description.
  • Turn an input image into a 3D scene, optionally guided by a prompt.
  • Explore a generated scene through a local browser-based viewer.
  • Save scene output as a Gaussian splat or a mesh in PLY format.
Before you choose

Requirements & limits

The project is available as a public GitHub repository under the Apache-2.0 license. The documented setup requires accepting the gated FLUX.1-dev model license.

  • The documented setup is Python-based and specifies Python 3.11, GPU-enabled PyTorch and additional model and 3D-processing dependencies.
  • The README suggests low_vram mode when GPU VRAM is below 24 GB.
  • The demo starts a local Viser server for viewing generated scenes; mesh display calls for the project’s customized Viser build.
  • ml-sharp support and background inpainting are labeled experimental in the README.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Generate and explore a scene with the demo

Run the documented demo to create a scene from a prompt or image and inspect it in a browser.

Before you start

  • WorldGen repository cloned recursively and installed in a Python 3.11 Conda environment.
  • PyTorch and torchvision with GPU support, plus the documented DA-2 and PyTorch3D dependencies.
  • For the documented gated model setup, accept the FLUX.1-dev license and log in to Hugging Face.

Steps

  1. Install the project and dependencies using the repository instructions.

  2. Run the command below for text input, or use the command below for image input.

    Documented command
    python demo.py -p "A beautiful landscape with a river and mountains"
    Documented command
    python demo.py -i "path/to/your/image.jpg" -p ""

  3. Open the local Viser server at http://localhost:8080 to explore the generated scene.

Expected outcome

A generated scene that can be explored in real time through the local Viser server.

Guide limitations

  • The README recommends setting low_vram to True when GPU VRAM is below 24 GB.
  • The --use_sharp generation path is labeled experimental.
  • The image demo example includes --use_sharp; the README labels that feature experimental.
Documentation 1 ↗
Source-based practical guide

Save a generated scene as a Gaussian splat or mesh

Use the Python API examples to generate a scene and save its output as a PLY file.

Before you start

  • WorldGen installed with its documented Python dependencies.
  • A Python environment with PyTorch available; the example selects CUDA when available and otherwise CPU.

Steps

  1. Import WorldGen and select a device with torch.device("cuda" if torch.cuda.is_available() else "cpu").

  2. For a Gaussian splat, create WorldGen(mode="t2s", device=device, low_vram=False), call generate_world with a text prompt, and save the result with splat.save("path/to/your/output.ply").

  3. For a mesh, call generate_world with return_mesh=True and save it using o3d.io.write_triangle_mesh("path/to/your/output.ply", mesh).

Expected outcome

A generated Gaussian-splat or mesh scene saved in PLY format.

Guide limitations

  • The README notes low_vram can be set to True when GPU VRAM is below 24 GB.
  • The mesh demo specifies a customized Viser installation for correct mesh visualization without backface culling.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
ZiYang-xie/WorldGen · Official homepage title · retained in its source language
How we review sourcesSuggest a correction