How to photograph an object for 3D: AI image-to-3D vs photogrammetry
For AI image-to-3D you need one great photo: a three-quarter view of the object alone, on a plain background, in soft even light. For photogrammetry you need many consistent photos: one every 10–15° around the object, in two or three height rings, with locked camera settings. Both reward the same habits, so the checklist below covers both.
- Light it softly from two sides through a diffuser. No flash, no direct sun.
- Plain, contrasting background. A curved poster-board sweep is ideal.
- Lock focus and exposure and keep the same zoom for every shot.
- AI: one sharp three-quarter view. Photogrammetry: 24–36 shots per ring, 2–3 rings, then the underside.
- Matte down anything shiny or clear before you start.
Photogrammetry vs AI image-to-3D
| AI image-to-3D | Photogrammetry | |
|---|---|---|
| Photos needed | 1 | Usually 50–150+ |
| Capture time | Seconds | 20–60 minutes |
| Hidden sides | Invented by the AI | Measured from your photos |
| True scale | No | Yes, with a scale reference in the shot |
| Works on drawings and old photos | Yes | No |
| Works on pets and people | Yes, from one quick shot | Hard: they move between shots |
| Fails on | Glass, chrome, cropped objects, busy backgrounds | Shiny, clear, featureless or moving subjects |
| Best for | Figurines, gifts, quick replicas | Accurate replicas, restoration, fit-critical shapes |
Short version: if the back side matters, use photogrammetry. If it just has to look right on a shelf, one photo and the AI 3D Model Generator is a lot less work. If you need exact dimensions for a part that fits something, measure it with calipers and use text-to-CAD.
Photo-shoot checklist for 3D
Tick these off as you set up. Your progress is saved in this browser.
- Object cleaned of dust and fingerprints, which show up as bumps.
- Shiny or clear areas matted with scanning spray, dry shampoo or cornstarch.
- Plain background in a color that contrasts with the object.
- Two soft lights at about 45° on each side, diffused, with no hard shadows.
- Flash, HDR, portrait mode and filters off.
- Main 1x lens, not the ultrawide.
- Focus and exposure locked so they don't drift between shots.
- Whole object in frame with a small margin, filling most of the picture.
- Scale reference placed (a ruler or a coin) if you need true size from photogrammetry.
- Test shot checked at full zoom for sharpness before you shoot the set.
Camera settings: phone or DSLR
A recent phone is plenty for both methods. Consistency matters more than megapixels.
- Phone: use the main camera at 1x. Tap and hold on the object to lock focus and exposure (Android "Pro" mode works too). Turn off HDR, night mode, portrait mode and beauty filters, which change the image from shot to shot.
- DSLR or mirrorless: manual exposure, a fixed focal length, aperture around f/8–f/16 for depth of field, ISO as low as your light allows, and a tripod if shutter speeds drop below about 1/60 s.
- Never change zoom mid-set. Photogrammetry software assumes the lens stays the same.
- Shoot RAW or high-quality JPEG and don't edit photos individually. If you correct them, apply the same correction to all.
Lighting for 3D scanning
Both AI and photogrammetry read shading as shape, so hard shadows and bright highlights turn into fake dents and flat spots. You want flat, boring, even light. It makes a dull photo and a great scan.
- Overcast daylight outdoors is a free giant softbox.
- Indoors, use two identical lamps with the same color temperature at about 45° on either side, both diffused.
- Fill the shadow side with a white card or foam board.
- Cross-polarization (a polarizing film on the lights and a polarizing filter on the lens, rotated 90°) removes most reflections from glossy objects.
Build a DIY lightbox in 15 minutes
- Get a cardboard box a bit bigger than your object in every direction.
- Cut windows in the left, right and top, leaving a 3–4 cm border.
- Tape baking paper or thin white fabric over the windows as diffusers.
- Lay a sheet of poster board inside so it curves from the floor up the back wall, with no corner line.
- Place one lamp outside each side window, plus one on top if you have it.
- Take a test shot. If you can see a distinct shadow, move the lamps closer or add more diffusion.
Background
A plain, matte background in a contrasting color lets the software separate the object from everything else. Use a light background for dark objects and a dark one for light objects. Avoid patterns, other objects, and glossy surfaces like a varnished desk, which reflect the object back.
One photo for AI image-to-3D: the hero shot
- Angle: a three-quarter view, 30–45° off the front and slightly above.
- Framing: the whole object in frame, filling most of it.
- Distance: step back and zoom in slightly rather than shooting close with a wide lens.
- One object only. Remove anything touching it.
- Pets and people: catch a natural, compact pose and use the dedicated pet or face generator.
More on what generators need in image-to-3D best practices.
How to capture photos for photogrammetry
Photogrammetry works by finding the same surface points in neighboring photos, so each photo should overlap the next by about 60–80%. Work in rings, and move the camera or the turntable by the same step every time.
- Low ring: camera at about 15–20° above the table. One shot every 10–15°, which is 24–36 shots for a full circle.
- Middle ring: raise the camera to about 40–45° and repeat.
- High ring: about 60–70°, looking down at the top. Fewer shots are fine here.
- Close-ups: add a few overlapping shots of areas with fine detail or deep recesses.
- Underside: flip the object and shoot another one or two rings so the software can join both halves.
- Scale reference: leave a ruler or a known-size object in some shots, or measure one dimension with calipers to scale the model afterwards.
- Check before you pack up: zoom into a few random photos. Re-shoot anything blurry now.
Turntable trap: if the object turns and the camera stays still, the background doesn't move with it. That confuses the software unless the background is completely featureless or you mask it out. Most photogrammetry apps can mask the background automatically or from a few hand-drawn masks.
You shoot one photogrammetry ring with a photo every 15°. How many photos is that?
360° ÷ 15° = 24 photos. At 10° steps you'd need 36. Two or three rings plus the underside puts a typical object at 70–150 photos.
Tricky objects and how to handle them
| Object | Problem | What to do |
|---|---|---|
| Shiny metal, glossy paint | Reflections move between shots | Scanning spray, heavy diffusion or cross-polarized light. |
| Glass, clear plastic | Software sees through it | Coat it matte. There's no reliable way around this. |
| Plain, featureless surfaces | Nothing for the software to match | Add temporary texture: small bits of masking tape, or a light dusting of spray. |
| Very dark objects | Detail lost in shadow | More light, a lighter background and slightly longer exposure. |
| Thin parts: wires, straps, antennae | Too thin to reconstruct | Remodel them in CAD and join them on afterwards. |
| Pets, people, plants in wind | They move | Use one photo and AI image-to-3D instead. |
| Large objects: furniture, cars | Too big for a turntable | Walk around it in rings, keeping the same distance. |
Tip: AESUB's vanishing sprays come in versions that evaporate on their own. AESUB Blue stays scannable for about 1–2 hours and is gone in roughly 4. Spray lightly, because thick coats hide fine detail.
After the scan: clean it up for printing
Photogrammetry meshes arrive heavy and noisy, often with a slab of table or turntable attached. Cut away the base, keep only the largest part, fill holes, smooth lightly and reduce the triangle count. The scan-cleanup recipe in Mesh Repair is Fix my model, then Make solid at Fine, a Light Smooth, and Simplify at Medium. Then follow making AI and scanned models printable.
Free PrintPal toolMesh Repair & SimplifierClean up a noisy scan: fill holes, drop the fragments and cut the triangle countOpen toolFrequently asked questions
What is the difference between photogrammetry and AI image-to-3D?
Photogrammetry measures an object from dozens of overlapping photos, so every side is real geometry. AI image-to-3D predicts a full model from one photo, which is far faster but invents the sides the camera did not see.
How many photos do I need for photogrammetry?
Shoot one photo every 10 to 15 degrees around the object, which is 24 to 36 per ring, and do two or three rings at different heights. A detailed object often takes 100 to 150 photos including the underside.
Can I use my phone to photograph objects for 3D?
Yes. Use the main 1x lens, lock focus and exposure, turn off portrait mode and filters, and keep the same zoom for every shot. Good light matters far more than the camera.
How do I scan shiny or transparent objects?
Make the surface temporarily matte with a vanishing 3D scanning spray, or dry shampoo or cornstarch on objects you can clean. Soft, heavily diffused light and cross-polarized lighting also cut reflections.
Related articles
Sources and further reading
- 3DF Zephyr: photography guide for photogrammetry (overlap, rings and camera settings)
- AESUB: vanishing 3D scanning sprays
- PrintPal: Mesh Repair & Simplifier
Tools for thisAll tools
Watch it doneVideo library
Thanks! We read every note and use them to decide what to update next.