Image to 3D Model: Turn One Picture into a 3D Splat (Free to Try)
Tutorial · 2026-11-04 · 6 min read · by SplatMart Team
Got one picture of an object? Here's how to turn an image into a 3D model online in 2026, which images work best, what AI can and can't recover from a single photo, and when to take more photos instead.
Image to 3D means handing an AI model one picture and getting a 3D object back. In 2026 that object is often a Gaussian splat: millions of soft, coloured points that look photographic from every angle and load straight in a browser. SplatMart's Pygmalion model does exactly this. Upload one image, or describe an object in words, and it generates a 3D splat you can spin, edit and download.
How to turn an image into a 3D model
- Open SplatMart Studio and sign up free. No card is needed.
- Choose Pygmalion and add one image of a single object, as a JPEG, PNG or WebP.
- Start the generation. The finished splat appears in your Studio.
- Spin it to check the sides your picture didn't show, then download it or open it in the SplatMart editor to crop and tidy it.
Pick an image that works
The model can only build on what it can see, so the picture matters more than any setting.
- The whole object is in the frame, with nothing cut off.
- The background is plain, so the object's outline is obvious.
- It's shot from a three-quarter angle, so the front, a side and the top all show.
- It's sharp and evenly lit, without hard shadows.
- Glass, mirrors and very thin objects are the hardest. Try something solid first.
What one image can and can't do
A single photo can't show the back of an object, so the model invents it. That's fine for game props, concept pieces and quick mock-ups, where a believable back is enough. It's the wrong tool when the result must match reality, like a specific product, a room or a property listing.
For a faithful copy of something real, take more pictures. Leighton 2 rebuilds real objects and whole spaces from a phone video or a set of photos, and our photos to 3D model guide walks through the capture.
One image or many photos?
- One image or a text prompt (Pygmalion): the quickest start and works from any picture, but the unseen sides are guessed. Single objects only, not rooms.
- A phone video or many photos (Leighton 2): true to the real thing and works for objects and whole spaces, but you need to walk around the subject while you capture.
Most other AI image-to-3D tools output meshes rather than splats. Our AI 3D model generator comparison covers how they differ.
Frequently asked questions
Can I turn an image into a 3D model for free?
Yes. SplatMart's free plan includes credits every month with no card required, enough for several Pygmalion generations from your own images.
What kind of image works best for image to 3D?
One object, fully in frame, on a plain background, shot from a three-quarter angle so the front, a side and the top all show. A sharp, evenly lit JPEG, PNG or WebP works best.
Can I make a 3D model of a room from one photo?
Not well. Pygmalion makes single objects, not scenes. For a room, record a slow phone video and use Leighton 2, which rebuilds the real space.
Will the 3D model match my object exactly?
The side in your picture will match, but the back is invented because one photo can't show it. Spin the result before you use it, or capture more photos with Leighton 2 for an exact copy.
What do I get at the end?
A 3D Gaussian splat that you can view in the browser, edit in the SplatMart editor and download.