Add a photo. See how a machine sees it.
An artificial intelligence model sees a photo in its own way. It calculates what's near and what's far, pixel by pixel. Here you can watch it work, on your own picture.
Try it
Your photo stays on your device. We don't receive it and we don't save it.
Getting the model means a download of about 24 MB, only now and only if you ask for it. On mobile data, Wi-Fi is a better idea.
Your photo, the depth map, and your photo moving with the computed depth.
Generated. Depth Anything V2 Small.
The depth was computed by a model, Depth Anything V2 Small.Ready-made example
The same test on one of our own images, for anyone who can't or would rather not use the model.
Generated. GPT image.
Depth mapGenerated. Depth Anything V2 Small.
The depth was computed by a model, Depth Anything V2 Small.What just happened
The model is called Depth Anything V2 Small. It looked at your photo and estimated the distance of every point. The lighter the area on the map, the closer it is.
It didn't work out who you are. It didn't recognise anyone. It made an estimate, and estimates can be wrong, especially with mirrors, glass and hair.
In a filmmaker's hands, the same tool can lift a face from its background or move the camera inside a frame.
Why we do this
A festival that asks people to show the trick starts with its own. This page is a small, real tool, and it says exactly what it is.
How the production journal worksThe machine does the sums. You make the film.