Can I turn a photo into a 3D model to print?

Two ways to get a 3D model from photos

• AI generators make a whole 3D model from a single photo. They guess the sides and back they can't see, so the result looks right but isn't measured. Good for figures, pets, toys and ornaments.
• Photogrammetry builds an accurate model from many photos taken all around a real object. Good when the shape and size matter: a part to copy, a sculpture, a scan of a room.

AI generators: one photo

Free tiers as of November 2026:

• Meshy: 100 credits a month free, and 10 downloads a month (its Lite model). Downloads include STL and GLB. Free results are under CC BY 4.0: you may sell prints if you credit Meshy.
• Tripo: a free monthly credit allowance; free results can't be used commercially.
• Hunyuan3D (Tencent): a free web app with daily generations, exporting OBJ and GLB. Its downloadable open model isn't licensed in the EU, the UK or South Korea.
• MakerLab (MakerWorld, Bambu Lab): Image to 3D with a choice of engines, a few free tries then credits, and a multi-colour 3MF for AMS printers. Its PrintU tool makes figures from portrait photos.
• TRELLIS.2 (Microsoft): open source (MIT). Running it yourself needs an NVIDIA graphics card with 24 GB of memory, but there's a free demo on Hugging Face. It outputs GLB.

The tips that make the most difference: a sharp, well-lit photo of the subject alone, on a plain background, seen from a three-quarter angle so some depth shows.

Photogrammetry: many photos

• KIRI Engine: phone app, free for scans of up to 150 photos, with unlimited STL, OBJ and GLB exports.
• RealityScan (Epic Games): free for individuals and for companies under $1 million revenue. The phone app takes up to 300 photos per scan; the desktop app (formerly RealityCapture) is the high-end option.
• Meshroom: free and open source for the desktop. It wants an NVIDIA graphics card; without one only a rough draft mode works.

For a good scan: walk all the way round the object, overlap each photo with the last by about three quarters, keep everything sharp, and make sure every part of the surface appears in at least two photos. Around 80 to 150 photos suits most objects. Shiny, see-through and plain single-colour surfaces scan badly.

Getting the model ready to print

Generators and scans make models to look at, not to print. Expect to fix a few things:

1. Convert it. Most give you GLB or OBJ. GLB to STL converts in your browser, sets the size in millimetres and stands the model upright.
2. Close the holes. AI models often have open edges, loose pieces and faces inside each other. Run them through Repair STL, or Fix Model in your slicer.
3. Set the size. The scale is arbitrary. Scale evenly to the height you want in Resize STL.
4. Thin parts. Hair, fingers, tails and swords can come out thinner than your nozzle. Anything under about 0.8 mm (two lines of a 0.4 mm nozzle) may not print; make the model bigger or thicken those parts.
5. A flat bottom. Feet and bases are rarely flat. Cut a little off the bottom with Merge and cut STL (a box set to cut away), or add a base there.
6. Too many triangles. Scans especially can run to millions: Reduce STL before the slicer struggles.
7. Colour. The colours of AI models are textures, which an STL can't hold. For a multi-colour print, use a generator that exports a multi-colour 3MF, or paint the model in your slicer.

For a flat picture, print it flat

A portrait, a landscape or a pet photo often makes a better print as a picture than as a statue:

• A lithophane: the photo appears when light shines through. One filament, and a classic gift.
• A filament painting (HueForge-style): the photo in a few colours of filament, swapped by height.
• For a logo, a drawing or clipart, Image to STL traces it into a sign, a keychain or a stencil. More in image or logo to STL.

Whose photo is it?

The generators' licences cover the model they make, not the photo you start from. Use your own photos, and ask before printing someone else's face.