DepthPop — turn a photo into an interactive spatial scene
DepthPop estimates depth from a single photograph and displaces it into a scene you can
explore by moving the pointer. Everything runs in your browser: your image is never
uploaded for processing.
How it works
- Choose a photo, or start from one of the bundled sample scenes.
-
Depth is estimated locally with Depth Anything V2 Small, using WebGPU where available
and WebAssembly otherwise.
- Refine the depth with a brush, set framing and fold, then export.
Privacy
Your photo is never uploaded. Depth is estimated on your device, in the browser. The only
requests DepthPop makes are for its own files and the depth model.
Browsers and offline use
It works in any modern browser with WebGL2. Chromium browsers use WebGPU when it is
available, and other browsers fall back to the slower WebAssembly. Safari and iPhone
performance have not been verified yet. The depth model is about 26 MB, downloads once, and
is kept in the browser, so DepthPop also works offline after the first visit.
What you can export
A preview image, a depth map, an editable project file, your original image, or a
five-second video of the Duo folding, in a format your browser supports. DepthPop is free
and needs no account.
What it is not
DepthPop produces a constrained 2.5D effect. It cannot recover parts of a scene hidden
behind objects, and it does not produce a metric 3D model.
Languages
The interface is available in English, Traditional and Simplified Chinese, Japanese,
Korean, Spanish, French, German and Portuguese.
A plain-text summary for language models is at /llms.txt.
DepthPop needs JavaScript and WebGL to run.