How to prompt an AI 3D model generator for correct shapes

AI 3D models can come out with mistakes, such as an extra leg or a second tail. This guide shows you how to write prompts that give the right number of legs, one tail, separate limbs and a clean outline. It also has prompts you can copy and tips for turning photos into 3D models.

Why AI 3D models sometimes go wrong

Meshsmith makes your model in two steps. First, the AI draws one picture of your object from your prompt. Then a second AI turns that picture into a 3D shape.

The second step only sees the front of the picture. It has to invent the back. If legs cross over, a tail curls behind the body or wings overlap, it has to guess. Those guesses cause most mistakes:

  • Extra legs, or legs that split into two.
  • Two tails, or a tail that grows out of the wrong place.
  • Wings fused to the body or to each other.
  • Missing parts, such as a leg, an ear or the back of a sword hilt.
  • Melted details: faces, fingers and small shapes that blur into lumps.

This happens with every AI 3D model generator, not only Meshsmith. The good news is that the way you write your prompt makes a big difference. The rules below fix most problems.

If a generation fails with an error, your credits are returned automatically. A model that finishes with the wrong shape still counts, so check the shape with a free draft first.

The core rules for AI 3D model prompts

Each rule has a prompt that works and one that tends to go wrong. You don't need to describe the background or lighting. Meshsmith already asks for a plain grey background, even lighting and one centred object.

Name exact counts

The AI does not know how many legs a dragon should have unless you say so.

Do

a red dragon, four legs, two wings, one tail

Don't

a red dragon with wings and legs

Keep every limb apart and visible

The 3D step sees one picture. Legs apart and a tail curving out to the side, clearly visible, give it nothing to guess. Meshsmith already frames every text prompt as one centred object, so you don't need to ask for a camera angle.

Do

legs apart, one long tail curving out to the side, clearly visible

Don't

a dragon coiled around a rock

Describe one object in a simple pose

One standing object is easy to turn into a solid shape. A group is not.

Do

a fox standing on all four legs

Don't

a fox and her cubs playing in the grass

Avoid action poses and overlapping limbs

Running, fighting and curled-up poses hide legs behind each other, so they come out merged or doubled.

Do

a wolf standing still, legs straight

Don't

a wolf leaping mid-air, howling

Describe material and colour simply

One or two materials and plain colour names are easy to follow.

Do

dark oak wood with brass corners

Don't

iridescent shifting rainbow ethereal glow

Say "game asset" or "collectible figurine"

These words ask for a clean, solid object with a clear outline.

Do

a knight, stylised game asset

Don't

a knight, cinematic, dramatic lighting

Keep it short: about 10–30 words

Long prompts bury the important parts. The AI may drop the counts you care about.

Do

a brown horse, four legs, one tail, standing, collectible figurine

Don't

a paragraph about the horse's history, the farm and the weather

Prompt templates you can copy

Copy a template into the studio and change the animal, object, colour or material. Keep the counts and the pose.

Creatures

Four-legged animal

a grey wolf, four legs apart, one bushy tail, standing, collectible figurine

Dragon

a red dragon, four legs, two wings, one long tail curving out to the side, standing, stylised game asset

Bird

a blue parrot, two legs, two wings folded at its sides, one tail, perched upright, game asset

Humanoid character

a dwarf warrior, two arms at his sides, two legs apart, standing straight, stylised game asset

Props

Chest

a wooden treasure chest, closed lid, dark oak planks, brass corners, game asset

Sword

a steel longsword, straight blade, leather-wrapped grip, gold crossguard, game asset

Potion

a round glass potion bottle, red liquid, cork stopper, collectible figurine

Vehicles

Car

a red vintage sports car, four wheels, closed roof, stylised game asset

Spaceship

a small white spaceship, two wings, one cockpit, low-poly game asset

Before and after: fixing bad prompts

Before

an epic dragon breathing fire over a castle

There are two objects (the dragon and the castle) plus fire, which is not solid. The pose is an action pose, so wings and legs overlap. The AI may give the dragon two tails or merge it with the castle.

After

a green dragon, four legs, two wings, one long tail curving out to the side, standing, stylised game asset

Before

a cute cat curled up asleep

A curled-up cat hides its legs and tail under its body. The 3D step has to guess the hidden parts, so you may get extra paws or a tail in two places.

After

a ginger cat, four legs, one tail, sitting upright, collectible figurine

Before

a horse galloping

In a gallop the legs cross over each other. The result often has three or five legs, or two legs fused into one.

After

a brown horse, four legs apart, one tail, standing still, collectible figurine

Before

a group of knights in a battle

A crowd cannot become one clean model. Arms, shields and swords from different knights get mixed together.

After

a knight in silver armour, holding a sword at his side, standing straight, stylised game asset

Image to 3D tips: photos that make good models

When you upload a photo, Meshsmith skips the drawing step and builds the 3D shape straight from your picture. A good photo gives a good shape.

  • Take the photo from the side or at three-quarters (halfway between the front and the side).
  • Make sure every leg, arm and tail is visible and not hidden behind another part.
  • Use a plain background, such as a wall or a sheet of paper. A PNG with the background removed is ideal.
  • Show one object only. Move other things out of the picture.
  • Use good, even light. Avoid strong shadows and bright reflections.
  • Keep the whole object in the frame. Don't crop off feet, ears, wing tips or handles.

If the shape still goes wrong

  1. Check the shape with a free draft first

    A draft is free and shows the shape only, without colours. If the legs, tail and wings look right, then make the textured model with credits.

  2. Change one thing at a time

    If the draft has two tails, add "one tail extended behind" and try again. Don't rewrite the whole prompt, or you won't know which change helped.

  3. Try a different variation number

    Open Advanced settings in the studio and type any number in Variation number. The same prompt with a new number gives a new result. When you get a good one, reuse that number to get a similar result again.

  4. Use a photo or sketch instead

    If a prompt keeps going wrong, upload a picture. A clear side-view photo or a simple drawing with every leg visible tells the AI exactly what shape you want.

Test your prompt with a free draft

Drafts are free and need no card. They show the shape in about 2 minutes, so you can check the legs, tail and wings before you make a textured model.

Frequently asked questions

Why does my AI 3D model have extra legs or two tails?

The AI first draws one picture of your object, then builds the 3D shape from it. It has to guess the back. If legs or tails overlap in the picture, it can guess wrong. Name exact counts, ask for the legs apart and the tail out to one side, and keep to one simple pose.

Do I get my credits back if a model comes out wrong?

If a generation fails with an error, your credits are returned automatically. A model that finishes with the wrong shape still counts, so check the shape with a free draft before you spend credits.

How long should my prompt be?

About 10–30 words. Say what the object is, how many legs, wings or tails it has, the pose, the material and colour, and a style such as "game asset".

Do I need to describe the background or lighting?

No. Meshsmith already asks for a plain grey background, soft even lighting and one centred object. Use your words to describe the object itself.

Is a photo better than a text prompt?

Often, yes. A clear photo of one object on a plain background, with every leg visible, removes the guesswork. It is a good choice when a text prompt keeps giving the wrong shape.

What is a variation number?

It is a setting under Advanced settings in the studio. Leave it empty for a new result each time. Type a number to try a different version of the same prompt, and reuse it to get a similar result again.