Image to 3D: turn a photo into a usable 3D model
Learn how image to 3D works, choose GLB or STL, and follow the right workflow for websites, games, product views or 3D printing.
Image to 3D uses a picture as a reference to generate a three-dimensional object that you can rotate, edit or export. AI generation estimates the shape, including surfaces hidden from the camera. A heightmap converter instead maps image brightness to surface height. Both create geometry, but they solve different problems.
The useful question is what you want to do with the model. A colored object on a website, a game prop and a printable figurine need different outputs and different checks. This guide helps you choose the workflow before spending time on generation.
Choose the result before the tool
| Your goal | Useful starting output | What to check next |
|---|---|---|
| Rotate an object on a website | GLB with suitable materials | Loading, camera framing and mobile performance |
| Make a game prop | Editable mesh, often GLB or another engine-supported format | Scale, topology, UVs, materials and collisions |
| Show an ecommerce product | GLB for a compatible web viewer | Accuracy against the real product and every variant |
| Print a figurine | STL or another slicer-supported mesh | Dimensions, connected surfaces, thin features and supports |
| Make a raised logo or sign | Extruded outline or relief mesh | Clean edges, closed shapes and a suitable base |
| Print a backlit photo | Lithophane mesh | Thickness mapping and the intended light source |
GLB is the binary form of glTF, a format designed for exchanging and delivering 3D scenes. It can carry geometry, materials, textures and animation. STL describes triangle surfaces and does not preserve those textures. Read GLB vs STL before converting a file you need in more than one workflow.
What AI infers from a single image
A photograph records appearance from one viewpoint. It does not contain a complete measurement of the object. An image-to-3D system uses its learned understanding of objects to estimate depth, volume and unseen surfaces.
For example, a three-quarter photo of a toy mug may reveal the opening, handle and one side. The model still has to infer the bottom, the far side and any hidden handle attachment. If the handle is mostly covered, the generated attachment may differ from the real object.
This is useful for creative assets because you can start with an image rather than construct every surface manually. For a replacement part that must fit a real mechanism, measure the important dimensions and use CAD or a calibrated scanning workflow. Making a generated object larger does not correct an incorrectly inferred hole or proportion.
AI generation, relief and photogrammetry
These methods are often grouped under “image to 3D,” although their inputs and results differ.
- AI generation: a single image or a few reference views guide a plausible full object. Useful for concept art, props and figurines.
- Heightmap relief: brightness becomes height above a base. Useful for terrain, plaques and images designed as depth maps.
- Silhouette extrusion: a closed outline gains thickness. Useful for badges, signs and simple logos.
- Photogrammetry: many overlapping photographs provide observations of the actual surface. Useful when you can photograph the physical object from enough angles.
A dark shadow in a portrait is not automatically a deep physical feature. That is why a brightness-based conversion can produce an interesting relief without producing a natural full head. The AI vs heightmap guide explains which route suits photos, logos and lithophanes.
A practical image-to-3D workflow
1. Prepare one readable subject
Choose a sharp image with an unobstructed silhouette. Keep handles, ears, feet and other important parts inside the frame. A simple background makes the subject easier to distinguish. Avoid a collage of inconsistent angles.
For a stylized game prop, a clear concept drawing can work as a reference. For a physical product, use your own photographs and retain measurements so that you can verify the result. See how to prepare an image for framing and input details.
2. Generate the shape
Upload the image to an image-to-3D service and select the mode appropriate to your goal. If you need a textured web asset, check that the selected mode includes texturing and allows the required download. Geometry generation and texture generation can be separate operations.
On Image to STL, start in the image workbench. Upload JPG, PNG or WebP, sign in, review the displayed generation cost and generate the model. The workbench focuses on inspecting geometry and preparing an STL export. Current access and credits are explained on the pricing page.
3. Inspect more than the front
Rotate through a full turn and look underneath. Check handles, openings, narrow gaps and separate pieces. A painted dark patch may look like an opening in a textured preview even when the geometry is closed.
Use a solid view to judge the silhouette and a wireframe or mesh view to inspect structure. If an important feature is wrong, change the reference, regenerate, or edit that region in a modeling tool. Keep the original result so you can compare revisions.
4. Export for the next application
For web presentation, retain GLB and its materials. For printing, export a mesh your slicer accepts and confirm the imported size. Image to STL lets you set a longest-side dimension in millimeters before STL export.
If you already have a GLB, the local GLB-to-STL tool provides a separate printing copy. Keep the GLB for textured use. Converting formats does not add missing detail or repair mesh defects.
5. Validate in the destination
A viewer is the start of validation. A website needs a real mobile loading test. A game needs an engine import and in-scene performance check. A print needs a slicer layer preview. An ecommerce model needs comparison with the actual SKU.
Follow the guide for your next step: website display, game assets, ecommerce models, or STL inspection.
Can you do image to 3D for free?
There are several kinds of “free”: a hosted generation allowance, a free preview, a downloadable community model and software you run on your own hardware. They do not provide the same thing.
For a useful free workflow, verify that you can download your own generated model in the required format, that the relevant model version is eligible, and that its license permits your intended use. A service may show a preview while charging for export. Open-source software still needs compute resources. Our free image-to-3D GLB guide explains how to evaluate those conditions.
Does one picture create a finished production asset?
It can create a useful starting model, but “finished” depends on the destination. A static display object may need only cleanup and lighter textures. An animated character may need retopology, rigging and skin weights. A figurine may need thicker small features and a base.
Assess the generated object against a short acceptance checklist for your task. That turns a vague judgment about whether a model looks good into a decision about whether you can actually use it.
Sources and next step
Technical references checked November 18, 2026: Khronos glTF overview, Tencent Hunyuan3D-2 documentation, and ImageToStl's heightmap explanation. Hunyuan's open-source documentation describes its own release; it is separate from the hosted generation service used by this site.
Ready to start with a picture? Open the image-to-3D workbench, then inspect the result for the application you plan to use.