Rodin Text-to-3D Generator
No photo, no reference, no problem. Describe what you want and Rodin builds the geometry and textures to match, in whatever style you name. Here's how to write a prompt that works, which 21 style presets you can call on, and when text-to-3D beats starting from an image.
Concept to 3D, no reference required
Rodin's text-to-3D generator is a purely language-driven pipeline: describe an object, and the model produces matching geometry, proportions, surface detail and PBR textures in the same pass - no source photo needed. That makes it the faster path for conceptual iteration, since you're not limited to what you happen to have a reference image of.
It runs on the same underlying Gen-2.5 model as Rodin's other generation modes, so effort tiers, export formats and downstream tools like Partial Edit all carry over. The short version of this feature lives on the Text to 3D product page; this guide goes deep on getting good results from a prompt alone.
Three steps, prompt to export
- Write your prompt. Describe the object, its material and its intended use. Start simple - add visual detail only when you want more control over the result.
- Generate. Pick an effort tier and, optionally, a style preset. Rodin produces geometry, proportions, surface detail and textures together from the description.
- Preview and export. Review the result, redo it for free if the read on your prompt was off, then export to whichever format your pipeline needs.
What actually belongs in a good prompt
Hyper3D's own guidance is simple: start with a simple description, and add detail only when you want more creative control. Material and style cues belong directly in the sentence rather than left to default settings - naming "ceramic," "brushed metal," "low-poly" or "hand-painted" in the prompt itself changes the result more reliably than any toggle.
Not every prompt needs all four parts - subject alone is enough to get a usable first pass. Add material and style once you know the direction, and reach for the detail clause only for things that matter for your use case (a specific pose, a closed vs. open state, a particular era). See the dedicated prompt-writing guide for a longer walkthrough with more examples.
Three prompts, three results
"violet crystal cluster on a rock base, glowing translucent gems, stylized fantasy asset"
Subject (crystal cluster) + material (translucent gems) + style (stylized fantasy) - no detail clause needed for a simple prop.
"medieval treasure chest, oak planks with brass straps and rivets, weathered PBR texture, game-ready prop"
All four parts present - this is the level of detail worth using once you know exactly what you want out.
"retro robot head, cream painted metal shell, glowing cyan visor, cute collectible toy style"
Style cue ("collectible toy") does a lot of work here - it tells Rodin the proportions should read as charming rather than mechanically accurate.
21 built-in styles
Naming a style in your prompt (or selecting its preset) steers proportions and surface treatment toward that look, from clean photoreal through heavily stylized:
A named style narrows the range of results, which makes it easier to get a consistent look across a batch of related assets - useful for prop sets and environment libraries.
Same five tiers as the rest of Rodin
| Tier | What it's for |
|---|---|
| Extreme-Low | Fastest output - exploring many prompt variations quickly |
| Low | Light meshes for background and secondary assets |
| Medium | Default, balanced tier for most production work |
| High | Denser meshes with crisper geometric detail, for hero props |
| Extreme-High | Maximum detail once a prompt is locked in |
The efficient pattern: explore several prompt variations at Extreme-Low or Low, pick the direction that reads right, then regenerate just that prompt at a higher tier for the final asset. Full timing detail is in the Gen-2.5 complete guide.
What you actually get back
Geometry and PBR textures (albedo, roughness, metalness) generate together in one pass - there's no separate texturing step. Since text-to-3D runs on the same Gen-2.5 model as image-to-3D, the same geometry controls apply: Quad topology for rigging and animation-friendly meshes, Triangle topology for smaller real-time-ready files, and a RAW ceiling past 10 million polygons at maximum effort if you need sculpting-reference detail. See the low-poly guide for bringing a dense result down to a game-ready count with Smart Low-Poly.
Because there's no photo constraining the result, a Bounding Box ControlNet is the most useful shape-guidance tool here - set rough proportions before generating so the output matches a target silhouette or scale, especially when an asset needs to fit a specific slot in a scene. Voxel and Point Cloud ControlNet matter less without a reference to guide from; full detail on all three is in the Gen-2.5 guide.
| Format | Typical use |
|---|---|
| GLB | Web, real-time engines, default export |
| OBJ | Universal static-mesh interchange |
| FBX | Game engines, DCC round-tripping |
| USDZ | AR/VR, Apple ecosystem previews |
| STL | 3D printing |
Full breakdown of what each format carries is on the export formats page.
Which one should you use?
| Situation | Better fit |
|---|---|
| You're starting from a concept with no existing reference | Text-to-3D |
| You want fast exploration across many variations | Text-to-3D |
| You have a real photo or reference of the object | Image-to-3D |
| The output needs to match a specific real-world item closely | Image-to-3D |
| You've found a direction you like and want to lock in detail | Text-to-3D to explore, then Partial Edit to refine |
See the Image-to-3D guide for the reference-driven side of Rodin, and Partial Edit for fixing one region of a result without a full regeneration.
What people build from a prompt alone
Concept exploration
Generate 5-10 variations of an idea before committing to one
Original IP & props
Objects with no real-world reference to photograph
Environment libraries
A consistent style preset across a whole batch of assets
Rapid prototyping
Blocking out a scene before final assets are ready
See a full production pipeline example in the game development workflow guide.
Text-to-3D walkthroughs
Every Rodin guide on rodin3ds.com
Text-to-3D & related features
Getting started
Text-to-3D: frequently asked questions
How does Rodin's text-to-3D generator work?+
You write a natural-language description of the object you want, Rodin generates geometry, proportions, surface detail and textures together from that description, then you preview and export the result. No reference image is required - it's a purely language-driven pipeline.
How do I write a good prompt for text-to-3D?+
Start with a simple description and add visual detail only when you want more control. Naming the subject, its material (ceramic, metal, fabric) and its style (realistic, low-poly, hand-painted) directly in the prompt works better than leaving those to default settings.
How many style presets does Rodin support?+
21 style presets, spanning realistic options like Photorealistic and Realistic through stylized looks like Anime, Cel-Shaded, Low Poly, Voxel, Isometric, Hand-Painted and Pixel Art, among others.
Should I use text-to-3D or image-to-3D?+
Use text-to-3D when you're starting from a concept with no existing reference and want to explore variations quickly. Use image-to-3D when you have an actual photo or reference and want the output to match it closely. Many workflows start with text-to-3D to explore, then refine a chosen direction with image-to-3D or Partial Edit.
Does text-to-3D generate textures too?+
Yes. Geometry and PBR textures are generated together in the same pass, based on the material and style cues in your prompt, so there's no separate texturing step before export.