Concept to 3D
Turn a Reference Image into a 3D Model
Use your sketch, concept art, photo or an image made in ChatGPT as a starting point. Generate and inspect a textured model, then refine it for your engine, avatar or print workflow.
Uploading your own image is free and skips image generation. 3D generation uses the displayed app token cost. A generated mesh is a starting point, not a guarantee of production readiness.
Prepare a useful reference
- One clear subject, fully in frame, against a simple background. Avoid text, collages and overlapping objects.
- For characters, use a neutral full-body T-pose or A-pose with separated arms and legs. Visible hands help with later rigging.
- Use only images you have permission to use. The generator must infer hidden surfaces; a front image does not specify every detail on the back.
From image to an exported model
- Create an asset in Asset Studio. Upload your reference to keep images, tasks, models and versions together. No image yet? Generate a reference from a description, after reviewing its payment source and cost. Asset Studio guide.
- Review the image and optional views. Choose Generate reference views if you need more angles. Check consistency of clothing, limbs and proportions. A single clean image can go straight to model generation; extra views are not mandatory.
- Generate the 3D model. Select your reference and choose Create 3D model. Review the model settings and displayed token cost, then start once. Inspect the task's result before choosing another paid generation.
- Inspect and refine. Rotate the model and check hidden surfaces, thin features, scale and textures. Open Mesh Editor for supported cleanup and export operations. Quad Remesh changes topology; inspect its result too.
- Rig only when needed. Static props skip this step. Humanoids intended for animation need a suitable rig. Clean up topology before using Auto-Rigger, then check bending.
- Export for the destination. Use FBX for ordinary Unity import, GLB for glTF-compatible tools such as Godot or Blender, and OBJ for static geometry. Stock Unity needs a glTF importer for GLB. VRChat requires a rigged humanoid and SDK validation; printing requires mesh preparation and a slicer-compatible format.
Choose the next workflow
Common questions
The back or hands look wrong. What now?
Inspect the reference for ambiguity, overlaps and cropped limbs. Improve the input or edit the mesh externally where needed. Generation does not reconstruct unseen details exactly, and a texture can hide malformed geometry.
Can my own API key pay for this?
A connected OpenAI, OpenRouter or Gemini API account can pay for compatible reference-image generation with 0 app tokens. Your provider bills separately. Image-to-3D compute still uses app tokens. Uploading an existing image avoids the image-generation charge. Accounts and costs.
Can I import the result directly into my game or slicer?
Inspect geometry, scale, materials and the target format first. Generated output may need topology, rigging or watertightness fixes. Use the destination-specific guides above rather than assuming a GLB fits every application.