Prompt Meshy for image-to-3D
Last updated: October 12, 2026
Meshy turns an approved character or prop render into a textured game mesh that stays close to the art, with PBR maps, a polygon budget and a rig-ready pose set in the form. Pick it when the design is signed off and the job is getting it into an engine.
Let the agent write it
You don't have to set any of this up yourself. The Layer agent knows how to prepare a source image and set up each model it runs, from the same guidance this page is built on. Attach the render in chat, name the model and say what the mesh is for, for example: Turn this into a 3D model with Meshy V7.1: a mobile game character I'll rig later, about 10,000 faces, PBR materials. The agent sets the pose, the budget and the materials, shows you its plan, and runs it when you approve. Use the rest of this guide when you want to prepare the source and settings yourself, and see Write a brief for the agent for how to brief it.
When to use it
Use Meshy for characters, creatures with a clear body, and props whose design is already settled in an image: heroes and NPCs you will rig, weapons, chests, furniture and pickups. It keeps the face, costume and colours of the source, builds a character in an A-pose or T-pose on request, and returns PBR maps for a lit engine.
It is weaker at glass, chrome and other transparent or mirror-like surfaces, at hair strands and thin wires, and at whole scenes. Build a scene from separate assets, and for a shape that only exists as words, see Choose a model for a text-to-3D model.
Versions on Layer
Meshy V7.1
The version this guide's examples were made with. Finer geometry than V7 and refined topology, up to four source images, Ultra detail, a Remesh switch and a Texture image to steer the surfaces.
Meshy V7
The same inputs and controls as V7.1, one release older. Keep one asset set on one version so the meshes match.
Meshy V6
Up to four source images, pose, polygon budget, quads and PBR materials, without Ultra detail, Remesh or Texture image.
How to write the prompt
On an image run, Meshy builds the mesh from the source image and the settings; the prompt box is not sent with it. So the prompt that matters is the one that makes the source. Ask for one subject, the whole of it, the view, the pose, a plain background and flat light:
Full-body 3D game character render of a young fire mage: red hooded cloak over a leather tunic, brown boots, a belt with small pouches, holding a short carved wooden staff topped with a glowing ember in his right hand. Front view, straight-on orthographic-style camera at chest height, the whole figure from hood to boots in frame with a little margin. Neutral standing pose, feet shoulder-width apart, arms relaxed and held slightly away from the body so the hands and staff do not overlap the torso. Stylised hand-painted game art look, clean readable shapes. Plain flat light-grey background, soft even studio lighting, no cast shadows, no props on the ground, no text.
That render is the source in the clip above. Everything Meshy cannot see it has to guess, and everything it can see it copies, shadows included, so the source should show the subject plainly rather than dramatically. Then the form does the rest: Pose, Face limit, PBR materials and Quads.
What works
Set the pose in the form
For a character you will rig, open Advanced and set Pose to A-pose or T-pose. Meshy rebuilds the stance whatever the source shows; writing "T-pose" somewhere is not reliable. We ran the render above both ways on the same seed. On Default the mesh copied the render: arms down, the staff gripped against the leg. On A-pose the arms came out to about 45 degrees with open hands, and the face, costume and colours stayed as drawn.
A clean source image
One subject, whole and uncropped, filling most of the frame, on a plain white, grey or transparent background, at 1024 px or more on the short side. For a busy image, remove the background first.
Flat, even light
Shadows and highlights in the source are baked into the texture as false surface detail. Ask for soft, even studio light and no cast shadows when you generate the source.
A front or three-quarter view
Use a front view for a character and a three-quarter view for a prop, so the front, a side and the top all show. Keep the perspective low ("orthographic-style camera", a long lens); a wide angle or a top-down view distorts the proportions.
Limbs apart, hands clear
Arms and legs that touch the body or each other fuse in the mesh, and small or hidden hands come back with fused or missing fingers. Ask for arms held away from the body and hands clearly visible.
More than one view
A single image leaves the back to guesswork. Add up to four views of the same subject to Source images: front, side, back and three-quarter, at the same scale and under the same light.
What to avoid
Props in a character's hands
A character built for a rig should be empty-handed. In our A-pose run the mage let go of the staff, which was left standing beside him as part of the same mesh, and in the default-pose run it stayed fused to his hand and leg. Generate weapons and tools as their own meshes and attach them after rigging.
Glass, chrome, hair and wires
Transparent and mirror-like surfaces, hair strands and thin wires reconstruct poorly from an image. Simplify them in the source (sculpted hair shapes, matte materials) or model them separately.
Dramatic key art as the source
A moody, heavily lit frame with a background makes a worse source than a plain render, because the model reads shadow as shape and keeps it in the texture. Make a plain version of the approved art first.
Remesh off for a game asset
On V7 and V7.1, turning Remesh off keeps the raw sculpt with millions of faces. Leave it on for anything going into an engine.
Settings that matter
Source images takes up to four views of the subject. Face limit sets the triangle budget, from 200 to 300,000; our 10,000 face runs came back at about 10,300 triangles. Quads builds the mesh from four-sided faces, easier to edit and to bend at the joints once rigged; triangles are fine for a static prop. Generate textures paints colour onto the mesh, and PBR materials adds metallic, roughness and normal maps so it reacts to light in an engine. Ultra detail (V7 and V7.1) sculpts finer surface detail and takes longer. Rigging adds a skeleton in the same run. Under Advanced, Pose sets A-pose or T-pose, and Seed repeats a result.
Prompts to start from
These make the source image; generate them with an image model, approve the still, then turn it into a 3D model.
Character source
Full-body 3D game character render of a young fire mage: red hooded cloak over a leather tunic, brown boots, a belt with small pouches, holding a short carved wooden staff topped with a glowing ember in his right hand. Front view, straight-on orthographic-style camera at chest height, the whole figure from hood to boots in frame with a little margin. Neutral standing pose, feet shoulder-width apart, arms relaxed and held slightly away from the body so the hands and staff do not overlap the torso. Stylised hand-painted game art look, clean readable shapes. Plain flat light-grey background, soft even studio lighting, no cast shadows, no props on the ground, no text.
For a character you will rig, drop the staff from the prompt and generate it on its own.
Prop source
3D game prop render of a single iron-banded wooden treasure chest, lid closed, chunky stylised proportions, worn oak planks, dark iron straps and corner caps, a heavy brass lock plate on the front. Three-quarter view from slightly above, so the front, one side and the lid are all visible, the whole chest in frame with a small margin. Stylised hand-painted game art look. Plain flat light-grey background, soft even studio lighting, no cast shadows, nothing else in the scene, no text.
Use it in Layer
Add the approved render to a session's board.
Click View on the image, then Turn into 3D model. Create mesh opens in the side panel with the image in Source images.
Click the model name at the top of the panel and choose Meshy V7.1. The source image stays in place.
Leave the prompt empty: on an image run, the image is the brief.
Turn on PBR materials, and set Face limit for your platform.
For a character you will rig, open Advanced and set Pose to A-pose.
Click Create 3D model. The cost shows on the button before you click. When the mesh lands on the board, click View to turn it in the viewer.
Learn more
- Image to 3D (Meshy): what makes a good source image, how multi-view input fills in the back, and when to use an A-pose or T-pose.
- How to get better image-to-3D results in Meshy (Meshy Help Center): a checklist for the source image and for multi-view sets.
- Fix character pose, face, and hand issues in Meshy (Meshy Help Center): why the pose setting beats describing a pose, and how to keep hands, faces and limbs clean.
- Image to 3D API (Meshy): every setting behind the form, with its range and default.
- Image to 3D model complete guide (Meshy): worked advice on views, lighting and the materials that reconstruct badly.