Generate from image is now live in Splat Labs Cloud. Open any project you can edit, pick a photo of the thing you want (a chair, a vase, a piece of equipment), mask it, click Generate. A minute later it shows up as a placed 3D object inside your Gaussian Splat — selectable, movable, walkable from every angle.
No external tools. No GLB pipeline.
Watch it in action
Photo → masked object → walkable 3D, all inside the Splat Labs viewer.
How it works
One button in the 3D Objects panel runs the whole pipeline:
Generate from image
We host the inference on our own GPU workers and wire the output straight into the viewer's object library and save flow. No exports, no uploads, no round-tripping through a separate tool.
How to try it
- Open a project in cloud.splatlabs.ai, load the splat, and switch to edit mode.
- Open the right-hand toolbar → 3D Objects → Library tab.
- Click Generate from image.
- Upload a photo of the object you want, paint a quick mask on it (one positive tap is usually enough; the in-browser segmenter does the rest), and submit.
- The job runs on our GPU workers. A pending row appears in the library while it processes — typically under a minute. When it's done, the new object lands in your library as a normal placed object.
- Use the Translate / Rotate / Scale gizmo to put it where it belongs, then Save.
It behaves like any other placed GLB after that — selectable, transformable, shareable in the link you send to your client.
Who can use it
Available under all Splat Labs Cloud plans. Open any project you can edit, click Generate from image, and you're in.
What it's good for
- Virtual staging. Stage a captured property with a specific chair, lamp, or rug from a single product photo. No furniture asset library to maintain.
- As-built documentation. Drop in a 3D version of a piece of equipment, a sign, or a fixture you're proposing to add — annotated against the real scan.
- Listing media. Replace empty-room awkwardness with a placed object that matches the brand of the listing.
- Quick concept presentation. Show two or three versions of the same room with different placed objects, on a phone, in a link.
The structural geometry of the scan stays exactly as captured. The object is generated and inserted on top — measured, walkable, and consistent from every viewpoint.
Related reading
- AI 3D Redesign: Transform Any Space With a Text Prompt — same AI credit pool, full-room redesigns instead of single objects.
- How to Upload, Translate, Rotate, Scale, and Add a Ground Plane — the gizmo and save flow for any placed object.
Questions about enterprise rollout, AI credit packs, or hardware capture? Contact us — we're happy to help.
