How to turn a 2D character image into a 3D model with AI
To turn a 2D character image into a 3D model with AI, give the generator one clear image of the character with the background removed and let it build a textured mesh you can download as GLB. In Ludo.ai's 3D Asset Generator that is the Image to 3D tab: add the image, check the cutout the app shows you before generating, set the triangle ceiling (1,000 to 200,000, default 50,000), the texture type (PBR, Color or None) and the resolution (1024 or 2048), and generate for 3 credits. No artwork yet? Text to 3D renders a concept image from a prompt and hands it to the same step. Either way you end with a textured GLB you can optimize, re-texture, then rig and animate.
Last updated September 2026
What you need
- One image of the character: a single subject, clearly lit, in a readable pose
- A Ludo.ai account with credits (3 for the model, 1 for an optimize pass, 2 for a re-texture)
1Choose the image
One subject, one view, as little ambiguity as possible. A character standing with arms slightly away from the body gives the generator more to read than arms folded into the torso or a weapon crossing the chest, because anything that overlaps has to be guessed. Concept art, a sprite, a sketch and a photo all work as input. No image yet? Start on the Text to 3D tab, which renders a concept image from your prompt and passes it into this same step.
2Add it in the Image to 3D tab
Open the 3D Asset Generator and go to Image to 3D. Upload the image or pick one from your favorites. The background has to go before generation, so the app cuts it out and shows you the result. Look at it properly: a missing hand, a chewed edge or a leftover artifact in the cutout is a missing hand, a chewed edge or an artifact in the mesh. The cutout can be opened in the image editor if it took something with it.
3Set the triangle budget and the textures
Open Advanced. Max triangles runs from 1,000 to 200,000, default 50,000. A hero character in a desktop game can afford the high end; forty background characters on a phone cannot. Adaptive decimation (0 to 1) simplifies the mesh below that ceiling where detail is not needed, and 0 turns it off. Type is PBR (base colour plus the maps an engine lights properly), Color (flat colour only) or None, and Resolution is 1024 or 2048.
4Generate the model
Run Create 3D Model. It costs 3 credits, held when the job is accepted and refunded if it fails or you cancel. The result opens in an interactive viewer with snapshots from several angles. Turn it before you judge it: the side the image never showed is the side to check first.
5Optimize or re-texture instead of regenerating
Two actions fix a model without paying for a new one. Optimize Model re-topologizes the mesh to a lower triangle target for 1 credit: the cheap route to a mobile-friendly version of a model you already like. Re-texture generates fresh textures on the same geometry for 2 credits: a second colourway, or a clean surface on a shape that is already right.
6Rig it, if the character has to move
A prop can ship as it is. A character needs a skeleton: click Rig, pick a rig type and the bone naming your engine expects, then animate it from a preset or a prompt. That is 1 credit for the rig and 0.2 per animation. The whole sequence is in the rig and animate guide.
7Download
GLB carries the mesh and its textures in one file, plus the skeleton and any saved clips once the model is rigged and animated, and it is the file to hand your engine. OBJ, STL, PLY and OFF export the geometry alone, with no textures. The engine import guide covers what to do with the GLB on the other side.
Specs
| Input | One image, by URL or upload; the background is removed and shown for review before generation |
|---|---|
| Max triangles | 1,000 to 200,000, default 50,000 |
| Adaptive decimation | 0 to 1; simplifies the mesh below the triangle ceiling without losing detail, 0 disables it |
| Texture type | PBR (default), Color or None |
| Texture resolution | 1024 or 2048, default 2048 |
| Credits | 3 per model, 1 per Optimize Model pass, 2 per re-texture; held on accept, refunded on failure or cancellation |
| Result | A downloadable GLB plus snapshot images rendered from several angles |
| Download formats | GLB with textures; OBJ, STL, PLY and OFF as static geometry without textures |
| Tabs | Text to 3D, Image to 3D, Texture Generation, Rig & Animate |
| API and MCP | create3DModel, POST /assets/3d-model, same parameters |
Tips
- Judge the cutout, not the photo. The mesh is built from what survives the background removal, so a missing hand in the cutout is a missing hand in the model.
- Readable poses convert best. Anything hidden behind the character in the source image is invented by the generator, and invented geometry is where the cleanup time goes.
- Set the triangle budget for the job, not for the preview. 50,000 is generous for a character that will be one of forty on screen, and Optimize Model is a cheaper second pass than a second generation.
- Choose PBR if the engine lights your scene, and Color for a flat or stylized look where lighting response would fight the art.
- Generate first, optimize second, rig last. Rigging is a separate step on the finished mesh, so nothing has to be redone if you decide to lighten the geometry before the character starts moving.
FAQ
How many images do I need?
One. The generator converts a single source image into a textured mesh and infers the sides the image does not show. That is why a clear front or three-quarter view with readable limbs gives a better result than a dramatic pose: less of the character has to be guessed.
Can I use a 2D game sprite as the input?
Yes. Concept art, a sprite, a sketch or a photo all work as long as the subject reads clearly once the background is removed. A tiny pixel-art sprite carries less detail than a full-resolution illustration, so expect a simpler mesh from it.
What if I do not have an image of the character yet?
Use the Text to 3D tab. It renders a concept image from your description and passes it into the same image-to-3D step, so the only difference is that the picture is generated rather than uploaded.
How do I control the polygon count?
Before generating, the Max triangles slider in Advanced sets a ceiling from 1,000 to 200,000 (default 50,000), and Adaptive decimation simplifies the result below it. After generating, Optimize Model re-topologizes a finished model to a lower target for 1 credit, which beats generating again for 3.
What does the model cost?
3 credits for the textured model. An Optimize Model pass is 1 credit and a re-texture is 2. Rigging is 1 more and each animation is 0.2, so a generated, rigged, animated character is about 4 credits in total.
Can I use the result in a commercial game?
Yes. Assets generated in Ludo.ai are licensed for use in your commercial games on every plan, including the free trial credits. The current terms are in the Terms of Service and EULA.
Can I do this through an API?
Yes. create3DModel (POST /assets/3d-model) takes the same image and the same mesh and texture options, and returns the GLB plus snapshot renders. It is available over the REST API and the Ludo MCP server, alongside the rigging and animation tools. See API & MCP.
Related
More guides for 3D Asset Generator
- How to auto-rig and animate a 3D character with AIAdd a skeleton to any model in about a minute, name the bones for your engine, then animate it from a preset or a prompt.
- How to import an AI-generated rigged GLB into Godot, Unity, Blender and UnrealWhat the GLB carries, how Godot, Unity, Blender and Unreal each read it, and where a conversion step is still needed.