Photographing objects for image-to-3D
Image-to-3D generators are deeply opinionated about the photos you feed them. Get the camera, lighting, and background right and even a 2018 phone produces clean printable meshes. Get them wrong and a 50 MP DSLR produces blobs. This guide is the practical setup: what gear is enough, the lightbox you can build from a cardboard box, and the photo-orbit pattern that consistently works.
Object on a turntable in front of a plain white background. Two diffused lights at 45° either side — or shoot under an overcast sky outside. Phone on a tripod or stack of books, in the 2x telephoto lens, framed so the object fills 70% of the frame. Spin the turntable, photo, repeat 8 times for a full orbit. Done.
Camera: what's enough
Phone
Any phone from the last 5 years is fine. Important nuances:
- Use the telephoto lens (2x, 3x, or 5x). The main "1x" lens on most phones is a wide-angle that distorts the closest parts of objects (a "wide-nose" effect on faces, "barrel" on flat fronts). Telephoto is closer to a portrait lens and gives flat, faithful proportions.
- Tap to focus on the object — phones default to centre-focus which often picks the background.
- Disable HDR. HDR composites multiple exposures and can blur fine edges. A single sharp exposure is what the model wants.
- Disable portrait/blur modes. The bokeh is fake; the AI can't distinguish blurred background from missing geometry.
- Lock exposure if your phone supports it (long-press the focus tap on iOS, AE Lock on Android). Stops the camera re-metering between rotations and giving each view a different brightness.
DSLR / mirrorless
- 50–100 mm prime or zoom range. The classic "portrait" focal length.
- Aperture f/8–f/11. Gets the entire object in focus front-to-back without diffraction softening.
- Manual mode, fixed white balance. Locks exposure and colour across all views. Auto modes can shift between orbit positions.
- ISO 100–400. Low noise. Use a tripod and let the shutter speed go long if needed.
- Shoot RAW or high-quality JPEG. Avoid heavy in-camera processing (sharpening, contrast) — you want a clean image, not a punchy one.
You do not need a DSLR. The most common limit on image-to-3D quality is lighting and background, not sensor size.
Lighting: the actual most important thing
Image-to-3D generators bake shadows into the mesh as geometry. A dark shadow on the underside of a chin becomes a literal indentation in the mesh. Goal: even, diffuse light from at least two directions.
Easy setups (in order of cost)
- Outside on an overcast day. The entire sky is one giant softbox. Genuinely the best free lighting on Earth for this purpose. Pick a spot out of any harsh shaded patches.
- Next to a north-facing window (in the Northern Hemisphere) with white paper or a white sheet as a fill bounce on the shadow side. South-facing window on the other side.
- Two desk lamps with paper diffusers. Tape a sheet of white paper or tracing paper a few cm in front of each bulb. Daylight-balanced (5000 K) bulbs only.
- LED video light panels. $20–$40 on Amazon. Pre-diffused, daylight-balanced, dimmable. Easiest predictable setup.
- Lightbox / softbox tent. A folding white-fabric cube with built-in LED strips. $25–$60. Best background and lighting combined in one purchase.
Build your own lightbox in 10 minutes
- Take a cardboard box (shoebox-sized for small objects, ~50 cm cube for medium). Cut large rectangular windows out of three sides and the top, leaving 5-cm borders for structural rigidity.
- Tape white paper or tissue paper across the cut windows from the outside. This is your diffuser.
- Line the inside with plain white paper or fabric, including the back wall, which is your seamless background.
- Sweep the floor up the back wall in a smooth curve (use a single sheet rolled gently) so there's no visible seam between floor and wall.
- Shine a light through each diffuser window. Two side lights, ideally one above. Daylight LED bulbs.
- Shoot through the front opening.
What to avoid
- Direct sunlight. Hardest light source available. Use an overcast day or move to shade.
- Single overhead light. Deep shadow under everything. Add a fill or move the light to the side.
- Mixed colour temperatures. Warm tungsten on one side and cool daylight on the other gives the texture a colour cast that confuses the model.
- Coloured walls reflecting onto the object. A red wall behind the camera will tint the entire front of a white object pink.
Background: clean and contrasting
See Image-to-3D best practices for the full background discussion. The short version:
- Plain white sheet, paper, or fabric is the default that almost always works.
- Solid mid-grey or any colour that contrasts with the object is the second choice — useful when the object is white.
- Avoid: wood grain, brick, patterned fabric, anything cluttered, the object's own shadow on a glossy table.
- Sweep the background from horizontal floor up to vertical wall in a smooth curve so there's no horizontal seam in the photo.
Framing the shot
- Object centred, filling 60–80% of the frame.
- Camera at the object's mid-height, not looking down or up at it. Get on the floor for short objects, get a stack of books for tall ones.
- Some empty space above and below the object — gives the segmenter room to work without cropping anything.
- Keep the object in the same frame position across all the orbit views. The model expects orbit, not "object jumping around the frame between shots".
- Square aspect ratio if your phone supports it. Most generators downsample to a square input anyway, so framing square saves a cropping step.
The 8-photo orbit (the workflow that just works)
Even if your generator only accepts a single image, taking the orbit and picking the best view is a good habit. Some generators accept 4 or even 8 views, which consistently outperforms single-image for asymmetric objects.
- Place the object on a turntable. A $10 lazy Susan from a kitchen store. In a pinch: a paper plate, a 12" pizza pan, or a stack of two paper plates with a marble between them.
- Mark the starting position with a piece of tape on the turntable edge, aligned with a reference point on the wall behind.
- Lock the camera in place. Tripod, mini tripod, stack of books, phone holder — anything so it doesn't move between shots. The camera stays still; the object rotates.
- Take the front view. This is your 0° reference.
- Rotate the turntable 45°, take the next photo. Repeat 6 more times for 8 total photos covering a full 360°.
- (Optional) Take a top-down view and a low-angle view if your generator supports more than 8 inputs.
- Pick your 1, 4, or 6 depending on what the generator accepts. Front + back + left + right is the universal good 4-pack.
Eight equal-spaced marks at 45° intervals around the turntable edge saves you eyeballing rotation amounts. Cheap and saves a re-shoot when one of your 8 turns out to be at the wrong angle.
Tricky objects and how to handle them
| Object type | Trick |
|---|---|
| Polished or chrome metal | Dust very lightly with cornstarch or matte spray (the kind sold for product photography). It's water-soluble and rinses off. |
| Glass, clear plastic, crystal | Same matting trick, or paint a temporary opaque coat (washable) so the silhouette reads. |
| Very dark / black object | Shoot on a white background, not black. Add extra fill light so the surface detail isn't crushed. |
| Hair, fur, fluffy textures | Brush or wet down the fur so the surface is "tight". True fluff is not currently reconstructable; you'll lose it either way. |
| Very small object (under 2 cm) | Use macro mode if your camera has it, and step back to use the telephoto lens at a working distance the lens can focus at. |
| Very large object (furniture, vehicle) | Use the telephoto from across the room. Most lightboxes won't work; an overcast day outside will. |
| Object with moving parts | Fix the moving parts in place with tape or putty before the orbit so the silhouette doesn't change between shots. |
| Object you can't move or rotate | Walk around it, keeping the same distance and height. This is harder to do consistently than spinning a turntable — expect a slightly worse result. |
Quick sanity check before uploading
- Are all the photos roughly the same brightness? If not, the model interprets the brightness shift as the object getting darker/lighter from different angles. Re-shoot with locked exposure.
- Is the object always in focus? Soft photos produce soft meshes.
- Is the background consistent across all photos? Same colour, same texture.
- Are there harsh shadows touching the object? They'll bake into the mesh.
- Is the object the right way up (gravity-correct) in every photo? Upside-down or sideways photos confuse the model.
Related articles
Further reading
- PrintPal docs — Image-to-CAD workflow
- PrintPal — AI 3D Generator (single and multi-image input)
- PrintPal — Pet 3D Generator (specialised pet-photo workflow)