Kept on this device, never uploaded.
Nothing leaves this device.
Light
Look · Soft key
Mood
Lamp
Shadow
Surface
Finish
Reel
Path
Motion
Length
Shape
Drag the picture to move the frame.
Maps
What this is
Carved Light reads the depth of your picture, works out which way every part of it faces, and lights it again with lights you can move. Shadows fall where the shape would throw them, edges catch rim light, and the light that was already in the picture is taken out first so the new light reads true.
How to use it
- Drop a picture on the canvas. It is read for depth and relit at once.
- Drag a light across the picture. Drag its ring to raise or lower it, or pick a look along the bottom.
- Save a PNG, or open Reel to record the light moving as an MP4.
Read the guide
Words: monocular depth · depth map · normal map · relighting
Common questions
Why does it say my pictures need WebGPU?
Reading depth runs a model on your graphics card through WebGPU, which current Chrome, Edge and Safari have. Some graphics cards cannot run this model even then, and some browsers cannot run its runtime; the page says which. Either way the sample still lights, and the light still falls on your own pictures, flat, without shape or shadows.
Why is the first picture slow?
The depth model downloads once, 47 MB, and is kept in the browser. After that a picture is read in a few seconds. Meanwhile the light falls on it flat, and it takes its shape after the first pass, before the fine detail is in.
Where do my pictures go?
Nowhere. They are read and lit in your browser. The only things it downloads are the sample, the runtime and the depth model; nothing is uploaded.
Whose depth model is it?
Depth Anything V2 Small, from the Depth Anything V2 paper by Lihe Yang and others (2024), converted to ONNX by the onnx-community on Hugging Face. It is under the Apache License 2.0; the notice and the licence text are at /lab/_carved/MODEL-LICENSE.txt, and beside the model file itself.