Find Shorts

Find Shorts

The Find Shorts step pulls any YouTube video into the editor by URL. This is step one of the clipper workflow. Long video in, short video variants out.

How it works

Paste a YouTube URL on the dashboard or inside the editor. Subedit hands the URL to our extractor worker. The worker uses yt-dlp to fetch the file directly from YouTube, then hands you back a temporary download URL.

Your browser fetches the bytes from the worker, not through us. This was a deliberate choice because at scale the bandwidth bill for routing video through our app would have killed the business before the first paying user.

So when you paste a URL the actual download is browser direct to worker. Subedit is just the orchestrator.

What works

Most public YouTube videos work. Long form podcasts, lectures, panel discussions, conference talks, public livestreams once they have processed. Anything you can open in a normal browser window without signing in usually pulls.

The cap is 4 GB on a single import. Anything larger needs to be trimmed at the source.

What does not work

A handful of cases fail.

1. Members only videos. The worker has no account on the channel so the gate blocks it.
2. Age gated videos. Same issue. The worker cannot prove an age.
3. Region locked videos. If YouTube blocks the video in the worker region the worker cannot pull it.
4. Private and unlisted videos that require sign in. Public unlisted with a link works. Private does not.
5. Recent uploads under a few minutes old where YouTube has not finalized the transcode.

If you hit one of these the worker returns an error and the editor will show the failure. Try a different source URL or download the file locally and drag it in.

When YouTube blocks the worker

YouTube runs bot detection on every download request. Our worker uses fresh cookies from a throwaway Google account, rotated about every two weeks, to look like a real browser. When cookies decay YouTube serves a challenge page and downloads start failing across the board.

You will see this as bot detection or unable to extract video info errors across many URLs at once. Not just one bad video, but a sudden run of failures.

When that happens it is on us to rotate the cookies. We watch for it. Usually the gap is under an hour.

A note on what we keep

The worker stores your downloaded video for fifteen minutes then deletes it. That window is long enough for the editor to grab the bytes and short enough that we are not running a video host. We do not keep your source video on our servers past that window.