Photogrammetry vs AI image to 3D: which one fits your project
One measures a real object from many photos; the other infers a whole model from one. They solve different problems.
By the threeD editorial team. Published . 6 min read.
Both techniques turn photos into 3D models, so they are easy to confuse. They work in opposite ways. Photogrammetry measures: it finds the same points in many overlapping photos and calculates where they are in space. AI image to 3D infers: a trained model looks at one or a few images and predicts a plausible complete shape, including the parts it cannot see.
Side by side
| Photogrammetry | AI image to 3D | |
|---|---|---|
| Input | Dozens to hundreds of overlapping photos of a real object | One to a few images; can be a drawing or a render |
| Needs the real object? | Yes | No |
| Shape accuracy | High: based on measurement | Approximate: plausible, not measured |
| Hidden sides | Missing unless photographed | Filled in by the model’s best guess |
| Typical output | Very dense mesh with baked lighting | Moderate mesh with a generated texture |
| Hard cases | Shiny, transparent and featureless surfaces | Fine text, thin parts, exact proportions |
When photogrammetry is the better choice
- You need an accurate digital copy of a specific real object: a museum artefact, a building, a product for measurement.
- Surface detail matters more than clean topology, as in scanned rocks, bark and ground textures for games.
- You can spend time on a careful shoot with consistent light and full coverage.
Expect cleanup: scans are dense and need remeshing, and shadows from the shoot are baked into the colour, which usually needs a de-lighting pass.
When AI image to 3D is the better choice
- The object does not exist yet: concept art, a sketch, a product idea.
- You only have one photo, such as a catalogue image.
- Speed matters more than exactness: blockouts, previs, background props.
- You need a lighter, cleaner starting mesh rather than millions of triangles.
The trade-off is accuracy. The result looks right but is not measured, so check proportions against the real thing before using it for anything where dimensions matter. The image to 3D page covers how to get the most from a single input.
Using both together
The approaches combine well. A common pattern is to scan a real object with photogrammetry for its shape, then use AI texturing to create clean, light-independent materials, or to generate variations of it. Another is to generate a quick AI model to plan a shoot, then scan only the objects that need real accuracy.
A one-line rule
If you need this exact object, measure it with photogrammetry. If you need an object like this, generate it with AI.
Choosing a good input photo for the AI route is covered in how to choose a reference image.