AI and 3D: transform a photo into a 3D model for augmented reality
AI 3D generation turns one or a few photos of a product into a textured 3D model in minutes, with no modelling skills required. The result works well for simple, opaque objects, in 3D and in augmented reality alike. But the AI invents the sides it cannot see: for perfect accuracy, choose scanning or modelling instead.
No commitment, no credit card required.
What is AI 3D generation?
AI 3D generation (also called image-to-3D) means creating a 3D model from ordinary images, without a scanner or modelling software. An artificial intelligence model, trained on a very large number of 3D objects, infers the object's full shape from what it sees in the photo, then applies textures to it. See the definition in the glossary.
Until recently, getting a 3D model meant either scanning the object or having a specialist model it. AI opens up a third way that is much faster and accessible to small businesses and retailers who already have photos of their products.
How does AI turn a photo into a 3D model?
Tools differ, but the process generally follows these steps:
- Isolating the object: the AI cuts out the product and ignores the background, which is why a simple background and a single object per photo help.
- Inferring the volume: from one view, or several complementary views, the model estimates the three-dimensional shape, including the sides that do not appear in the photo.
- Rebuilding the mesh: the estimated shape is converted into a polygon mesh, the 3D surface that can be displayed.
- Texturing: the photo's colours are projected onto the mesh, and the AI fills in the areas it cannot see.
- Export: the model is delivered as a GLB file. It is often very dense, sometimes over a million triangles, so it has to be lightened for mobile augmented reality.
What are the limits of AI 3D generation?
AI produces impressive results on simple objects, but it has well-known weaknesses. It is worth keeping them in mind before choosing this method:
- Hidden sides are invented: with a single front photo, the back and underside are inferred, not observed. The back of an armchair or of a device may therefore differ from the real product.
- Transparent or reflective materials: glass, chrome, mirrors or glossy plastic fool the AI, which mistakes reflections for texture.
- Thin parts: chair legs, cables, handles, spokes or grilles often come out thickened, merged or missing.
- Text and logos: labels, brand names and lettering usually come out distorted or illegible.
- Dimensions: a photo carries no scale. The generated model will not have the product's real size until you set it.
- File size: the mesh is dense and the UVs are fragmented. The model displays fine in 3D but must be lightened for augmented reality on smartphones.
AI, 3D scanning or custom modelling: which method should you choose?
| Criterion | AI (photo to 3D) | 3D scanning (photogrammetry, LiDAR) | Custom modelling |
|---|---|---|---|
| Turnaround | A few minutes | Capture then processing: from a few tens of minutes to a few hours | From a few hours to several days depending on complexity |
| Accuracy | Good on the photographed side, approximate elsewhere | Very good on matte, textured objects | Highest, detail by detail |
| Effort and skills | Very low: one to four photos | Moderate: around a hundred photos and a careful method | High: a professional 3D modeller |
| Physical object required | No, photos are enough | Yes | No: drawings, dimensions or photos |
| Best for | Large catalogues, simple objects, first trials | Existing products with rich materials (wood, fabric, stone) | Flagship products, technical items, products not yet manufactured, variants |
These methods combine very well: AI to cover a catalogue quickly, scanning or modelling for strategic products. Our guide How to create a 3D model for augmented reality covers each of them in detail, and Visuality also offers custom modelling on quotation.
How should you photograph a product for AI?
The quality of the model depends directly on the photo. Check these points before starting the generation:
- ✅ A single object in the photo
- ✅ Object well centered and fully visible
- ✅ Simple / uncluttered background (ideally white or uniform, but not mandatory)
- ✅ Adequate but not excessive lighting
- ✅ A sharp photo, taken at the object's height, without a strong wide-angle effect
- ✅ In detailed mode: four views (front, left, right, back) taken from the same distance and under the same lighting
Avoid:
- ⚠️ Highly visible text in the photo (logo, label, sign, etc.)
- ⚠️ Multiple objects in the photo
- ⚠️ Complex / cluttered background (many elements behind the object)
- ⚠️ Object partially cropped or too small in the image
- ⚠️ Hard shadows and strong reflections on the object
How do you generate a 3D model with AI in Visuality?
- In your 3D object library, click Add an Object, then My Object is Not Modeled and Convert a photo into a 3D object.
- Give the object a title and choose the mode: simple generation (1 photo, ideally from the front) for a quick result, or detailed generation (4 photos: front, left, right, back) for an often more accurate result.
- Upload the photos from your computer, or from your phone by scanning the QR code shown on screen.
- Confirm: the generation costs 1 credit, whichever mode you choose. The model is ready within a few minutes and appears automatically in your library.
- Set the object's size to match its real dimensions, choose floor or wall placement, then share it by link, QR code or iframe.
For iPhone and iPad, Visuality then automatically generates the model's USDZ version, lightened if necessary, so that it opens smoothly in AR Quick Look. The illustrated step-by-step guide is available in the documentation on AI 3D generation.
How many AI generations are included in Visuality plans?
Every plan includes a monthly allowance of AI credits, topped up at each subscription renewal. Unused monthly credits do not roll over to the next month. If you need more generations, extra AI credit packs are available: they are used once your monthly credits run out. Each generated model also counts towards your plan's object limit.
| Plan | 3D Objects | AI credits per month |
|---|---|---|
| Bronze | 10 | 5 |
| Silver | 25 | 10 |
| Gold | 50 | 15 |