AI depth estimation for video
Drop in a video and get a depth map back, one for every frame: brighter is nearer the camera, darker is further away. The motion, camera move, framing and timing all survive; colour and texture do not. That is the point — it gives a video model, a compositor or a 3D tool the structure of real footage without dragging its appearance along.
8 credits per second of video
Up to 100 MB per clip
The depth video opens in a session you can come back to, ready to download or carry into another tool.
How it works
Drop in an MP4, MOV or WebM up to 100 MB. Resolution and frame rate follow the source, so nothing is resized or resampled on the way in.
The model reads the whole clip rather than each frame alone, and writes out how far every pixel sits from the camera. Because it looks at neighbouring frames, a wall that is not moving keeps the same depth instead of shifting from frame to frame.
The depth video opens in a session you can come back to, so you can download it, run the next clip, or carry the result into the workflow editor as the control input for a video model.
Before and after
What it is for
Feed the depth pass to a video model as a control input and it will follow your footage's layout, camera move and timing while inventing an entirely different look. Depth carries the geometry without carrying the faces, colours or textures of the original.
A depth pass drives depth-of-field, atmospheric haze, fog and relighting in After Effects, Nuke, Fusion or Resolve. It is the matte you would otherwise roto by hand, and it updates itself as the shot moves.
Displace a plane by the depth map and a flat clip gains real parallax — the 2.5D camera move, a spatial-video conversion, a wiggle-stereo pass. Depth is the one channel those effects all need and no camera records.
Because depth is continuous rather than a hard cutout, you can threshold it anywhere: isolate the foreground subject, drop only the background, or grade the middle distance on its own without keying on colour.
How the estimation works
Running an image depth model over a clip frame by frame is what makes depth boil: every frame gets its own independent guess, and the guesses disagree slightly, so flat surfaces crawl and edges buzz. A video depth model reads neighbouring frames together, so a surface that is not moving holds the same value and the pass stays usable in an edit.
The output is relative: it tells you what is nearer and what is further within a shot, not how far anything is in real units, and the scale is re-fitted per clip. That is what ControlNet, displacement and depth-of-field all want. It is not what a measuring tool wants, and no single-camera model can give you that.
Greyscale is the interchange format: every tool that consumes a depth map reads brightness, so that is the default. The four colour ramps — turbo, inferno, magma and viridis — spread the same values across a hue range, which makes the depth ordering far easier to read on screen but has to be converted back before anything downstream can use it.
Every plan includes credits you can spend on this page or anywhere else on YouArt, and a new account starts with free trial credits.
For hobbyists and explorers
For creators and pro users
For power users and teams
For teams and studios
Questions
More tools
Each one does a single job on a file you already have — no prompt to write and no editor to learn.
Beyond one pass
This page runs one fixed step. Open the workflow editor to wire the depth pass straight into a video model as a control input, run a batch of clips through the same graph, or chain it after an upscale.
