All toolssmart crop video API

Smart Crop Video to Vertical with an API

Reframe landscape video to 9:16, 1:1 or 4:5 by following the people in it. Faces are tracked, the active speaker is picked out from lips and audio, and the crop window holds still, pans or cuts the way an editor would. Run the preset directly on this page or call the same agent-ready endpoint from your product.

LIVE TOOL
Run smart crop video
Drop a video, adjust the defaults if you need to, and download the result here.

or ·

Already have an account? Sign in

BEFORE / AFTER

See what the tool changes.

A real example: the after side is the unedited output of a KinoPipe job. Run the live form above to generate the same result from your own media.

BeforeSource video
WebM · VP9
mountain.webm
AfterCompatible MP4
Ready
MP4 · H.264
mountain.mp4

Demo footage: public-domain timelapses (Bureau of Land Management, Oregon · NASA SVS). Processed by the same pipeline the API and MCP tools call.

Useful defaults, typed options.

The tool slug stays stable while your agent supplies named media inputs and a narrow set of documented options.

  • Follows whoever is speaking
  • Stacks two speakers who cannot share the frame
  • Blurred fit for shots without a face
What this preset does
  1. 01Detect the shots, so the window never pans across a cut
  2. 02Track faces 8 times a second and tell who is speaking from lips and audio
  3. 03Hold the window still on the speaker, pan when they move, cut to whoever speaks next
  4. 04Encode a 1080x1920 MP4 (1080x1080 for 1:1, 1080x1350 for 4:5)
Stable endpointPOST /api/v1/tools/smart-crop-video
Live API request
{
  "inputs": [{ "id": "main", "url": "https://example.com/podcast.mp4" }],
  "options": { "aspect_ratio": "9:16", "layout": "auto" }
}
Successful output example
{
  "kind": "smart_crop",
  "output": { "width": 1080, "height": 1920 },
  "shots": 23,
  "faces": 26,
  "speakerSwitches": 4,
  "seconds": { "fill": 89.6, "split": 0, "fit": 0.4, "center": 0 },
  "segments": [
    { "start": 0, "end": 11.04, "layout": "fill", "faces": 1, "reason": "one face" },
    { "start": 63.76, "end": 77.78, "layout": "fill", "faces": 2, "reason": "follow speaker" }
  ]
}

Know the boundaries before you run.

Made for footage of people: podcasts, interviews, talking heads, vlogs. Sports and screen recordings have no face to follow and fall back to a blurred fit.
Faces smaller than 5% of the frame height, or on screen for less than half a second, are ignored.
Speaker detection needs an audio track; without one the window follows the most prominent face.
Sources up to 30 minutes; trim longer recordings first (find-highlights picks the clips).

About smart crop

How is this different from video-to-vertical and crop-video?

video-to-vertical keeps the whole picture and pads it over a blur. crop-video cuts a fixed window (left, centre or right). Smart crop moves the window: it finds the faces in each shot, works out who is talking, frames that person and cuts to the next one when they start speaking.

What happens with two people on screen?

If both faces fit in the vertical window they are framed together. If they are too far apart, the window follows whoever is speaking, and when both speak during the shot the two faces are stacked top and bottom. That is layout auto; fill always follows one person, split forces the stacked view on every two-person shot.

What does result.analysis contain?

The list of shots with the layout chosen for each (fill, split, fit or center), how many faces were found, how many times the window cut to another speaker and the seconds spent in each layout, so an agent can check the result or retry with another layout.

What if a shot has no face?

It is fitted whole over a blurred copy of itself, the same look as video-to-vertical. Set no_face to center to crop the middle of the frame instead, which suits B-roll and screen recordings.

Related media tools