Text to 3D: Turn a Prompt into an Explorable 3D Scene
WorldGen turns a written description into a 360 degree panorama and then into a scene you can walk through: no modelling, no GPU, and no export pipeline to babysit.
What text to 3D means here
Text to 3D is the umbrella term for tools that turn a written description into a three-dimensional result, and the results differ enormously depending on what a tool is built to produce. Some generators return a single object: a chair, a helmet, a creature on a pedestal, ready to be dropped into a scene someone else has already built.
WorldGen returns the place itself. The output is an environment with a floor, walls or terrain and a horizon that stays coherent in every direction, so the camera moves instead of orbiting a prop. If the goal is to stand somewhere and look around, that difference decides which kind of tool the job needs.
How a prompt becomes a scene
The pipeline runs in two stages. A diffusion model first writes a 360 degree equirectangular panorama from your prompt. Because the whole field of view is decided in one image, walls meet their neighbours, the sky lines up overhead and the floor closes behind you before any geometry exists.
The second stage lifts that panorama into 3D: a Gaussian splat when photorealism matters, or a mesh when the result has to stay editable. Both come out of the same panorama, so the room you explore matches the image you were shown, and the space holds together as you turn around.
Working in two stages is also why generation is measured in seconds rather than minutes: the heavy reconstruction runs on hosted GPUs, not on your machine.
Writing a prompt that produces a usable scene
The model has to infer layout, light and materials from your sentence, so a prompt that names those things beats a prompt that names a mood.
- Lead with the place, not the feeling: "an abandoned greenhouse at dusk" tells the model far more than "something beautiful".
- Say where the camera stands: eye level, drone height, from the doorway, close to the water.
- Describe the light: golden hour, overcast, neon signage, shafts of sunlight through water. Light carries atmosphere better than adjectives do.
- Name materials: wet asphalt, rusted iron, polished concrete, moss-covered stone, weathered timber.
- Add at most one or two constraints: "no people", "clear sky", "photorealistic". Long negative lists dilute the prompt.
- Keep it short, around 20 to 40 words. The second stage needs room to interpret rather than repeat.
Where text to 3D scenes are useful
Scene generation shortens the distance between an idea and something a colleague can walk through. Teams use it for previsualisation and pitch decks, level and set concepts, virtual tours, property walkthroughs, exhibition staging and look development, where a prompt is faster than blocking out geometry by hand.
It is also a good first step for asset work: generate the environment from text, pick the frame that reads best, and use that frame as the reference for a more controlled pass.
What you get at the end
Every scene stays in your workspace as an interactive result you can open in the browser and share. When you are done exploring, export it as 3D Gaussian splats for photoreal playback or as a mesh for a modelling, lighting or rendering pipeline. Credits cover text and image input alike, so switching between the two costs nothing extra.
From prompt to scene in four steps
The same flow runs for every scene, whether you start from words or from a photo.
- 01
Describe the place
Write one or two sentences covering the location, the light and where the camera stands.
- 02
Generate the panorama
The first stage returns a 360 degree image in seconds. If the layout is wrong, adjust the prompt and run it again before spending time on geometry.
- 03
Convert to 3D
The second stage turns the panorama into Gaussian splats or a mesh while keeping the look you approved.
- 04
Explore and export
Walk the scene in the browser, then export splats or a mesh, or share the interactive link.
Output formats at a glance
The same scene can leave the generator in three shapes.
Frequently asked questions
Do I need 3D modelling experience?
No. You write a prompt and the scene arrives ready to explore. Exporting a mesh is optional, and nothing in the workflow requires a modelling tool.
How long does a text to 3D scene take?
The panorama stage usually finishes in seconds, and the 3D stage takes seconds to a couple of minutes depending on the queue and the detail you ask for.
Can I use generated scenes commercially?
You own the scenes you generate, subject to the terms of service. You stay responsible for the prompts you write and for making sure any reference image you upload is yours to use.
Related guides
Turn your next idea into a place you can walk through
Start with a sentence. New accounts include welcome credits, so the first scenes are on us.