Text to 3D: describe an object, get a mesh
Write a sentence about what you need. threeD returns four untextured preview meshes to choose from, then refines your pick into a fully textured model you can download or keep editing.
Sample prompt
a wooden lamp post with a hanging lantern, weathered timber, iron bracket, glass, cracked and sun-bleached, realistic, game-ready
- Subject
- a wooden lamp post with a hanging lantern
- Materials
- weathered timber, iron bracket, glass
- Condition
- cracked and sun-bleached
- Style
- realistic, game-ready
The structure is what matters: name the object first, then what it is made of, then how used it looks. Style words come last because they change the least.
How text to 3D works
- 1
Write the prompt
Name the object, its materials and its condition. Add an optional negative prompt for things to avoid, such as “no base” or “no text”. - 2
Pick a preview
Four untextured meshes come back for a low credit cost. Rotate them, compare silhouettes, and keep the one with the right shape. - 3
Refine and export
Refinement adds detail and PBR textures. Then remesh, rig or export straight away in the format your tool expects.
Why you choose a shape before you pay for texture
Most of what makes a generated model usable is decided by its silhouette: proportions, how many parts it has, whether thin pieces are connected. Texture is easier to change later. That is why generation is split in two.
The preview stage costs 10 credits and returns four candidates. Refinement costs 25 and only runs on the one you pick. Over a session this is cheaper than generating fully textured models and throwing most of them away, and it puts your judgement where it matters most.
What a prompt that works looks like
Short and concrete beats long and poetic. The generator understands objects and materials far better than moods or stories. A reliable order is subject, materials, condition, then style, as in the demo above.
- Describe one object. “A table with four chairs” usually merges into a single blob.
- Say what each part is made of. Material words drive both shape and the texture pass.
- Use the negative prompt for recurring problems: “no pedestal”, “no background”, “no text”.
- Ask for a pose explicitly for characters. A T-pose or A-pose makes auto-rigging work much better.
The longer text-to-3D prompt guide has before-and-after examples and a checklist.
Settings you can change before generating
| Setting | Options | When to change it |
|---|---|---|
| Art style | Realistic, stylised, low-poly, voxel | Match the rest of your scene. Low-poly also gives you a lighter mesh. |
| Target polycount | Adaptive, 3k to 300k triangles | Set it up front if you already know the budget for your platform. |
| Topology | Triangles or quads | Quads if you plan to sculpt, subdivide or rig the model. |
| Symmetry | Auto, on, off | Force it on for vehicles, furniture and characters. |
| Texture resolution | 1K, 2K, 4K | 4K for close-up hero assets; 1K is enough for most props. |
What to check before you use the model
Generated models are a strong start, not a guaranteed finished asset. Spend a minute on these checks before you build on one:
- Scale. Models come out normalised to roughly one metre. Set real dimensions before printing or placing in AR.
- Hidden faces. Look underneath and behind. Areas the generator could not “see” get less detail.
- Polycount. If it is heavier than your budget, run it through remesh rather than regenerating.
- Texture seams. If a seam shows, regenerate only the texture with AI texturing.
Questions people ask
How much does text to 3D cost?
Can it make characters?
Does it work in languages other than English?
Why does my model have a base or pedestal I did not ask for?
How is this different from image to 3D?
Make your first model in the studio
Every plan includes all the tools, commercial rights and every export format. Cancel any time.