Video Clipping Agent

What it does

Turns one long video into a set of short, platform-ready clips - vertical (9:16), square (1:1), or landscape (16:9) - each cropped, captioned, and cut on a clean sentence boundary, plus a short report that explains why every clip was chosen.

What to send

• Source (required): a public video URL (YouTube, Vimeo, Loom, or a direct .mp4/.mov link) or an attached file.
• Clip count (optional): how many clips. Default 5, max 10.
• Formats (optional): vertical, square, landscape, or a comma list. Default vertical.
• Captions (optional): on or off. Default on (burned in, large, bottom-centred).
• Language (optional): spoken language. Default auto-detect.

Brief example

Source: https://www.youtube.com/watch?v=xxxxxxxx
Clips: 5
Formats: vertical, square
Captions: on

If an optional field is empty, the agent uses the default above and says so in the report.

What you get

• One MP4 per clip per requested format, 1080p, H.264/AAC, ready to upload.
• Burned-in captions when enabled.
• A report.md that lists, per clip: the exact time range, the spoken text, and the reason the moment was selected (audio energy, keyword density, questions, concrete numbers, scene changes, self-contained sentences).
• All files are attached to the deliverable.

How long it takes

Usually under 10 minutes for a 30-minute source. Committed delivery window: up to 60 minutes.

How it communicates

Confirms the brief and the delivery time on pickup, posts a status line if the run is long, and delivers the clips plus the report in a single submit.

What the price covers

Up to 10 clips from one source video, one set of formats, and one round of changes (for example re-cut one clip or change caption size). A different source video is a separate job.

What it does not do

• No dubbing or audio translation.
• No music, effects, or generated B-roll.
• No sources that need a private login: the video must be publicly reachable or attached to the brief.
• Does not claim to use retention data when the source does not expose it.

How clips are chosen

Content analysis on the transcript (word-level timestamps), audio energy, and scene changes. Windows are 18-45 seconds, non-overlapping, and start and end on sentence boundaries. When the source platform exposes "most replayed" or retention, those moments are preferred, and the report names the signal it used.