Skip to main content

A ranked list of clips from one long video

You provide the video and choose the length and clip count. VimeoClip reads the whole transcript, proposes complete moments and shows why each one ranks where it does.

Podcast host at the mic · Illustration
Podcast host at the micIllustration
Illustration (stock video), not VimeoClip output · Storyblocks

From a long landscape video to a vertical clip that stands on its own

VimeoClip reads the whole transcript, anchors each clip’s start and end to spoken lines, prefers to cut at the end of a sentence, then reframes to vertical. The pictures below illustrate the reframing step with stock video; they are not VimeoClip output.

Talking at the mic · 16:9 · Illustration

[ 16:9 → 9:16 ]

Illustration (stock video), not VimeoClip output · Storyblocks

Fill 9:16 · Layout illustration
9:16Layout illustration

Layout illustration made from stock video · Illustration (stock video), not VimeoClip output · Storyblocks

  1. Transcribe word by word
  2. Pick complete thoughts
  3. Reframe to vertical, add captions

Layout illustration made from stock video: the vertical frame on the right is cropped from the landscape video on the left. This example did not go through VimeoClip’s selection or captions.

How moments are chosen

It reads the whole video first

Whisper transcribes word by word, PySceneDetect finds scene boundaries, and the system outlines the whole video before it scores any part.

Cuts follow the speech

The model returns opening and closing quotes rather than seconds. Code anchors those quotes to real transcript words, then prefers to end on a finished sentence.

  • Where captions carry little punctuation, as in Vietnamese, a pause of 0.45 seconds counts as a sentence boundary
  • A clip does not open on a connector such as "and", "so" or "then"

Ask for the part you want

Type a topic or a question to narrow the selection. The system checks that request against the transcript.

Scores come with reasons

Each clip scores on four axes: Hook, Flow, Value and Spread. Measured defects, such as an approximate cut or an ending mid-thought, cost points and get a label.

Options for each job

OptionValues
SourceUploaded file, YouTube, public links yt-dlp can read
Clip lengthAutomatic up to 3 minutes, up to 5 minutes, under 90 seconds, or fixed ranges
Aspect ratio9:16, 4:5, 1:1, 16:9; one per job
LayoutFollow the speaker, two people stacked, screen, or plain crop

Private videos that need a sign-in usually cannot be fetched from a link. Upload the file instead.

Common questions

No. The score sets your review order and does not predict views. Review clips before you post.

Try it on your latest long video