Generate a 3D Model from Photos: A Single-Image Workflow

Generating a 3D model from photos usually means one of two very different jobs. Multi-photo photogrammetry solves camera positions and measures geometry across many images. Single-image reconstruction - what TripoSplat does - infers a plausible 3D scene from one picture. This page is about that second job: how to generate a 3D result from a photo here, what to expect from the output, and how to keep the source from limiting the result.

Single image, not a photo set

The hosted lane takes one image. That is the whole input, so the practical question is not "how many photos should I take" but "which single photo reconstructs best". A clean, evenly lit, single-subject photo beats a large set of inconsistent ones every time, because a set does not help a lane that only reads one frame.

If you have a photo set and you need measured geometry, that is photogrammetry, and it is a different tool. If you have one good photo and you want a viewable 3D asset quickly, this lane is built for exactly that.

Choosing and preparing the photo

1. One subject. Isolate it. Props, hands, and background clutter get reconstructed alongside the object.
2. Fully in frame, not cropped. Anything you cut off is guessed as cut off.
3. Even, diffuse light. Strong reflections, glare, and hard specular highlights become false surface detail.
4. Readable edges. A subject that contrasts with its background reconstructs a cleaner silhouette than one that disappears into a similar colour.
5. Fixed before upload. Straighten and crop with the free image cropper, and shrink oversized files with the image compressor. Both run locally in the browser.

Uploads accept JPEG, PNG, WebP, GIF, and AVIF up to 10 MB.

Running the generation

1. Open the AI 3D Generator.
2. Upload your prepared photo or choose a sample.
3. Set the Gaussian budget, inference steps, guidance scale, and output format. PLY is the default; SPLAT is available when a viewer expects the compact format.
4. Run the job. A guest can try one lightweight preview first; a signed-in TripoSplat run is priced 48-72 credits, shown before you submit.
5. Download the PLY or SPLAT file and inspect it with the free PLY and splat viewer.

Reading the output honestly

The result is a Gaussian splat, which is a scene you look at, not a solid you measure. Two things to check before you rely on it:

• Coverage. Rotate the view. Faces that were never visible in the photo are inferred.
• Scale. There is no real-world unit baked in. Decide the intended size downstream, for example with the model scale calculator.

If your downstream task needs polygons, see the 3D mesh generator guide for the mesh lane, and the image to 3D print guide if the goal is a physical part.

Common mistakes

FAQ

Can TripoSplat use multiple photos at once?

No. The lane takes one image. For multi-photo, measured reconstruction you need a photogrammetry tool.

What comes out of a generation?

A Gaussian splat file, PLY or SPLAT, which you view and download from the workspace.

Do I need to sign in?

Guests can run one lightweight preview. To generate at full quality and keep the file in account history, sign in - new accounts get 60 one-time credits valid for 30 days.

How long does a photo take to become a model?

It depends on the provider queue and the settings you choose. The workspace polls the job and shows progress until the file is ready.

Start from the AI 3D Generator, or read the image to Gaussian splat workflow for the input checklist in more detail.