Documentation
Everything the studio does, written down.
A sentence goes in, a 3D model comes out, and all of it happens inside your browser. These pages say exactly what happens in between.
Getting started
Open the studio, type one sentence such as an old rusty lantern, and press Generate. The model appears in the viewer a moment later and is saved to the Library in this browser. Download it as GLB or OBJ from the Files tab.
There is nothing to install and no account. The first visit downloads the studio and the engine (about one megabyte, compressed); after that they are cached and work offline.
Quality tiers
The studio picks a tier for your device when it opens; you can change it at any time.
| Tier | Triangle budget | Textures | Typical time on a laptop |
|---|---|---|---|
| Draft | 800 | colours only | under a second |
| Low | 2,500 | colours only | about a second |
| Medium | 12,000 | 256 px painted maps | one to six seconds |
| High | 45,000 | 384 px painted maps | a few seconds to half a minute |
Draft and low files are usually 10 to 200 KB, which suits websites and prototypes. Medium and high files are larger (often 0.5 to 4 MB) because of the painted textures. Phones and devices with 4 GB of memory or less start on low; desktops start on medium; high is a choice you make.
Every request has a time limit and a Cancel button. If a request runs out of time the studio says so and suggests a lighter tier.
Writing prompts
Name the thing first, then describe it. The engine knows 266 things by name across nature, containers, weapons, furniture, buildings, characters, vehicles, sci-fi, food and anatomy, plus 47 assembled recipes.
- Colour and material words are honoured. "a black iron lantern", "a golden goblet", "a red sports car".
- States change the look. old, rusty, mossy, snowy, broken, glowing.
- Sentences compose. "a table with a candle on it", "a knight holding a sword", "a treasure chest full of gold".
- Counts and shapes. "three barrels" makes a cluster; "a teapot shaped like a dragon" bends one thing into another.
- Size in metres if it matters: "a 2 m statue". Otherwise the typical real size is used.
- Style comes from the Style control or from words such as "low poly", "pixel art" or "cute".
If a word is unknown, you still get a model: the engine builds the closest form it can and the Details tab says what it assumed.
What comes out
- GLB, the main format: one file with geometry, PBR materials and, from medium up, embedded PNG textures. Each file passes a structural check (header, buffers, accessors, materials) before you see it.
- OBJ and MTL for older tools. OBJ carries geometry and colours only; textures stay in the GLB.
- Real-world scale in metres, Y up, the pivot on the ground at the centre.
Privacy
The studio is a set of static files. Generation runs in a background thread of the page, the Library is stored in your browser (IndexedDB), and downloads are made from memory. No prompt, model or identifier is sent anywhere. There is no account, no cookie and no analytics.
To remove everything, delete an asset from the Edit tab, or clear this site's data in your browser settings.
Devices
The engine is plain JavaScript running on the processor, so it needs no graphics card and no special browser features. The viewer uses WebGL, which every current browser has; without it the studio still generates and downloads files and says the preview is unavailable.
Pictures
Turning a photo into a 3D model needs picture and shape models that are gigabytes in size. The on-device studio builds from words, which is fast on every device. Describe what is in the picture instead: "a blue ceramic vase with a narrow neck" works well.
Using a model
GLB opens in Blender, Godot, Unity (with a glTF importer), Unreal, three.js, Babylon.js and most web viewers. Check the scale, pivot and materials against your project's conventions.
Inside Privera, the toolbox's 3D Asset Studio uses the same engine and can send a model straight into a vibecoding project, where a single line shows it on a page: <div data-vib-model="assets/barrel.glb" data-autorotate style="height:420px"></div>
Licences and commercial use
Models made by the on-device engine are yours, including for commercial use. The engine, its GLB writer and its PNG encoder are Plasmated's own code; the viewer uses three.js (MIT).
The showcase and gallery models on this site were made with the full Plasmated pipeline, which used these open parts:
| Part | Licence | Note |
|---|---|---|
| Picture model (FLUX.2 klein) | Apache-2.0 | commercial use permitted |
| Shape model (Hunyuan3D-2 mini) | Tencent community licence | commercial use permitted below 100 million monthly users; not licensed for use in the EU, the UK or South Korea |
| Super-resolution (Real-ESRGAN) | BSD-3 | |
| Unwrapper (xatlas), CLIP scoring (open_clip), validator, three.js | MIT / Apache-2.0 | |
| Human base bodies (MakeHuman) | CC0 | |
| Female base body by Gelado1 | CC BY | attribution: Gelado1 |
| Environment map | CC0 |
This page is a summary, not legal advice.
FAQ
Why is my model simpler than the showcase?
The showcase pieces came from the full pipeline with large picture and shape models. The on-device engine builds procedurally in seconds; try medium or high quality and a more specific sentence.
Where is my library stored?
In this browser on this device. Another browser or device has its own library.
Does it work offline?
Yes, after the first visit.
Is the same sentence always the same model?
With the same seed, yes. Leave the seed on auto, or press New, for a different take.