Tutorials · Published on 2 October 2026 · By the usi3d team

Text to 3D: how to describe your part so the AI models it well

For the AI to model your part well, say what the object is, its shape and proportions, the details it needs, its style, its material and colour, and what it is for. One to three specific sentences are enough: “a faceted low-poly desk owl in amber PLA” works far better than “a nice owl”. Anything you leave out, the AI decides for you.

Typing “a dragon” and getting a dragon is easy. Getting the dragon you pictured, sitting down, chubby and small enough for the shelf, depends on how you put it. Text-to-3D AI asks no questions: it takes your words and fills every gap with the most likely answer. This guide covers what is worth saying, what is not, and when text is the wrong starting point.

How the AI reads a description

Think of the AI as a fast sculptor who has never seen your idea. Every specific word closes a door. “Planter” leaves a thousand shapes open; “faceted geometric planter with its saucer” leaves a handful. Vague adjectives such as “nice”, “cool” or “modern” tell it almost nothing, because they mean something different to everyone.

Order helps too. Start with the object and its overall shape, move on to the details and finish with style and material. Cram three ideas into one long sentence and the AI tends to latch on to the first.

The formula: what to put in a text-to-3D description

Put together: “A headphone stand with a round swivel base and a channel for the cable, in matte coral PLA.” You do not need every part every time. Object, shape and a couple of details already make a visible difference.

Good and bad prompt examples

What to avoid when describing a part

• Scenes. “A knight fighting a dragon in a castle” is three objects and a background. Ask for one part; if you want the group, make it piece by piece.
• Negatives. “No wings” sometimes brings wings, because the word is in the sentence. Describe what you do want: “a dragon with its arms close to its body”.
• Text and lettering. An engraved name or a sign will not come out reliably legible. Add it later in your design software or slicer.
• Fine dimensions. “A 3.2 mm hole for an M3 screw” will not be respected. If the part has to fit another, that is a job for CAD.

Ask for shapes that print well

While you are describing, describe something printable. Ask for a flat base or a plinth (“sitting on a round base”), compact bodies and chunky parts. Outstretched wings or a thin sword in hand will need supports and may snap when you remove them. With a 0.4 mm nozzle, the thinnest parts should be at least 1.2 mm thick at the scale you intend to print.

When to use text and when to use an image

Text works when the idea is yours and no photo exists: an invented character, a planter with a particular shape, a decorative object. If the object exists, a photo is usually more faithful; we explain it in how to turn a photo into a 3D model. And if you have a pencil sketch, upload it to Image to 3D: a clear outline conveys proportions that are hard to put into words.

How to create a model from text on usi3d

1. Open Text to 3D

   Go to Text to 3D or, if you would rather be guided through a conversation, the AI agent. If you are not sure where to start, try one of the ideas on the page and change it to suit you.

2. Type or dictate the description

   Follow the formula: object, shape, details, style, material and colour. You can dictate it aloud with the microphone and add reference images.

3. Check the cost and confirm

   Preparing the quote uses no credits. You see the cost before confirming and, if the generation fails, the credits come back.

4. Review the model

   Turn it round in the viewer. Check thin parts, overhangs and the base: whatever the text did not say, the AI has made up.

5. Carry on with your part

   Every version stays in My projects. From there you can paint it in Colours, cut it up with Split into parts or download it as STL or GLB.

How to fix a description that did not work

If the result is not right, do not rewrite the lot. Work out what went wrong and change only that: each new generation uses credits, and this way you learn which word made the difference. A few examples:

• Wrong proportions: add a comparison. “Big head, half the size of the body.”
• Too much detail: ask for “simple shapes, smooth surfaces”.
• Too plain: name the missing details, one by one.
• Wrong pose: describe it with verbs. “Sitting, with its tail curled round its feet.”

After a few tries you will learn which words the AI respects and which it ignores. If this is your first time on usi3d, our launch post maps out all the tools.

Try it on usi3d: Text to 3D