Photogrammetry vs AI photo to 3D
Both produce a 3D model. They ask for completely different things to get there. Photogrammetry measures an object you have, from many photos. We infer one from a single image in about 5 minutes. Here is the honest split, including where photogrammetry is the better tool.
Side by side
| Photogrammetry | 2dto3D (one photo) | |
|---|---|---|
| Input | Dozens to hundreds of overlapping photos of an object you physically have. | One clear photo. It can be a supplier packshot of something you have never held. |
| Setup | Turntable or a full walk-around, even lighting, a matte non-reflective surface. | None. Upload the image from wherever it already lives. |
| Time per object | Capture plus reconstruction, typically an hour or more per object. | About 5 minutes, unattended. |
| Real-world size | Measured, if you put a scale reference in the frame. That is its strongest claim. | Estimated from the object we identify. Close on everyday products, less reliable on replicas and collectibles. |
| Surface accuracy | Captures the exact object, including wear, print, and imperfections. | Reconstructs a faithful version of the object. It is realistic, not forensic. |
| Glossy, dark, transparent items | Its known weak spot. Reflective and clear surfaces confuse the solver. | Handled, because the model is built from generated views rather than from reflections. |
| Cost per model | Rig, lighting, software, and staff time. The per-object cost is mostly labour. | $2 per model, from $5 up front. No subscription. |
| Output | A dense mesh you usually retopologise and bake before shipping it anywhere. | A web-ready GLB plus AR, an embeddable viewer, and GLTF, FBX, OBJ, USDZ, STL, and 3MF. |
They are not doing the same job
Photogrammetry is a measurement technique. It finds the same surface point in image after image and solves for the shape that explains all of them. Every requirement it has follows from that: many overlapping photos, steady lighting, a surface that looks the same from every angle, and the physical object in front of a camera. When those hold, it reproduces that specific object, marks and all.
We start from the opposite constraint: one photo, and usually no access to the object at all. So instead of matching views, we generate the views the photo is missing, then build a textured mesh from the full set. The result is a realistic model of the product rather than a survey of one physical copy of it. That is the trade, and it is the right one when the model is going on a product page, into a game engine, or into AR in a shopper's room.
If you want the mechanics rather than the comparison, read how it works or the generation walkthrough. The same pipeline is described for the search terms people arrive on: image to 3D, photo to 3D model, and 2D to 3D converter.
Use photogrammetry when
Dimensions have to be exact
Engineering, restoration, replacement parts, anything where a measurement is the deliverable. Put a ruler in frame and photogrammetry measures it.
The specific object matters
A one-off artefact, a used item, a piece with markings or damage that has to survive into the model. You are documenting this object, not a category of object.
Use one photo when
You only have the photo
A supplier packshot, a catalogue image, a product that ships from a warehouse you have never visited. There is nothing to walk around with a camera.
You need a catalogue, not a centrepiece
Hundreds of listings, each worth a few minutes and a couple of dollars. A capture rig does not scale to a catalogue, and most of a catalogue does not justify one.
Plenty of teams use both: photogrammetry for the hero product, one photo for the other four hundred listings. See what the single-photo path produces by industry.
Photogrammetry vs AI: common questions
What is the difference between photogrammetry and AI photo to 3D?
Photogrammetry measures. It matches features across dozens or hundreds of overlapping photos of a real object and solves for the geometry those photos imply, so it needs the object, a capture setup, and time. AI photo to 3D infers. From one photo it generates the views the photo is missing, then builds a textured mesh from all of them. Photogrammetry reproduces the exact object you photographed; AI reconstructs a realistic version of it from a single image.
Is photogrammetry more accurate than AI photo to 3D?
For dimensional accuracy, yes. With a scale reference in the frame, photogrammetry measures the object and an AI model estimates it. If a number on a spec sheet has to be right, use photogrammetry. For visual realism on a product page, the gap is much smaller, and the input requirement usually decides it: photogrammetry needs the object in front of a camera, and most catalogue work does not have that.
Can I use one photo instead of photogrammetry?
Yes, if what you need is a model people look at rather than measure. Upload one clear, well-lit photo of a single object and you get a textured 3D model in about 5 minutes that shoppers can spin, explore, and place in their room in AR. You do not need a turntable, a light box, or the object itself.
Does photogrammetry work on shiny or transparent products?
Poorly, and this is its best-known limitation. The technique relies on matching the same surface point across photos, and reflective, dark, or clear surfaces do not hold still between angles. Glassware, polished jewellery, and chrome typically need matting spray or a different technique. Generating the views instead of matching them sidesteps that problem.
How many photos does photogrammetry need?
Usually 50 to 200 overlapping images for a small object, shot so that every surface appears in several frames with consistent lighting. A single photo, or four photos from the front, back, and sides, is not enough for a photogrammetry solver. It is enough for us, because we generate the intermediate views rather than matching them.
Can I 3D print a model made from one photo?
You can export STL or 3MF and print it, and it will look like the object. Treat the physical size as an estimate rather than a measurement: we infer real-world scale from the object we identify, so check the dimensions before printing anything that has to fit something else.
Models built from a single photo
See the full gallery
See what one photo gets you
No rig, no turntable, no object required. Built for realism, ready in about 5 minutes, and you start from $5, pay as you go.