AI 3D Scene Generator for Text and Images
WorldGen is an AI 3D scene generator that turns a text prompt or one reference image into a 360 degree panorama and then into a walkable scene. You can explore the result in a browser and export it as 3D Gaussian splats or a mesh.
Last verified 2026-10-20
What is an AI 3D scene generator?
An AI 3D scene generator creates a navigable environment rather than a single object. The deliverable has spatial continuity: floors, walls or terrain connect around the viewer, the horizon closes behind the camera, and the user can move through the result instead of orbiting a prop.
That distinction matters when comparing 3D tools. Meshy, Tripo and Hyper3D Rodin focus on individual assets for a production pipeline. WorldGen focuses on the place itself, beginning with a text prompt or one image and ending with an explorable scene.
How does WorldGen turn a prompt into a walkable scene?
WorldGen uses the two-stage pipeline described by the WorldGen research project. Stage one generates a 360 degree equirectangular panorama, which fixes the overall layout, lighting and appearance before geometry exists. Stage two estimates depth and lifts the panorama into 3D.
The same look can be delivered as 3D Gaussian splats for photoreal playback or as a mesh for modelling, lighting and rendering workflows. Working in two stages is what keeps the room or landscape coherent when the camera turns around.
Can you start with text, an image, or both?
Use text when the place does not exist yet. A useful prompt names the location, camera height, lighting and materials; it does not need to describe every object. Use an image when composition, palette or lens already matters and you want the surrounding world to grow from that reference.
The inputs can also be combined in sequence. Generate a scene from text, choose the frame with the strongest composition, and use that image as a new reference when you want tighter control over the next iteration.
What can you do with the generated 3D scene?
Open the result in the browser to review sight lines, scale and atmosphere with another person. Share the interactive scene, or take it into a downstream workflow as Gaussian splats or a mesh. The right format depends on whether the next step is playback or editing.
For game environment concepts, set design, film previsualisation and property walkthroughs, the value is speed: one prompt or image produces a place that can be evaluated before a team commits to modelling or final art.
From prompt or photo to a scene you can explore
Every route uses the same two-stage pipeline and the same browser workspace.
- 01
Choose the input
Describe the location, or upload one image whose framing and visual direction you want to keep.
- 02
Approve the panorama
Review the 360 degree image before conversion. Change the prompt or reference if the layout is wrong.
- 03
Convert to 3D
The approved panorama becomes Gaussian splats or a mesh while preserving its layout and look.
- 04
Explore and export
Walk through the environment, share the interactive result, or export it for the next tool.
Scene output by job to be done
WorldGen delivers the environment as an interactive scene and can export it in two production formats.
Frequently asked questions
Do I need a GPU or 3D modelling software?
No. Generation and browser exploration run as a hosted web workflow. A modelling tool is only needed if you choose to continue editing an exported mesh.
Is this a 3D model generator or a scene generator?
WorldGen is built for scenes. It produces an environment with spatial continuity, not a single character or prop on a neutral background.
Can I use both text and images?
Yes. Both inputs feed the same two-stage scene pipeline, so you can choose the material that best defines the place.
What can I export?
The pipeline supports interactive scenes, 3D Gaussian splats and mesh output. Export availability follows your current plan and the options shown in the product.
Key takeaways
- Choose a scene generator when the environment, camera movement and 360 degree continuity are the deliverable.
- WorldGen fixes layout in a panorama before converting the scene to Gaussian splats or a mesh.
- Text is best for inventing a place; an image is strongest when the reference already defines the look.
Related guides
Sources and verification notes
Product capabilities are based on the official pages listed below and were verified on 2026-10-20. These links describe the products named on this page; they are not endorsements.
Turn a prompt or photo into a place you can walk through
Generate the panorama, approve the composition, and explore the finished environment in the browser.