Long video → captioned Shorts

Give AI a long video and an editorial brief, then get a pack of selected, reframed, and captioned social clips.

Recipes

Run it

  • Video
  • 5 pipelines
  • ≥ 2 credits

AI Agent

Terminal

New here? Quick start
01
Homebrew brew install pipe2-ai/tap/pipe2
↳
Go go install github.com/pipe2-ai/pipe2-cli/cmd/pipe2@latest
02
Get an API token ↗ echo "$PAT" | pipe2 auth login --token -

3 clips from one episode

Change the result

--inputrequired
asset_urldefault—Source video: a YouTube or social URL, direct media URL, local file, or existing Pipe2 asset.
--instructions
stringdefaultFind the five strongest self-contained moments. Prefer clear ideas, surprising claims, useful explanations, or complete stories. Each clip should make sense without the rest of the video.Editorial brief describing which moments to select and what makes a good clip.
--max-clips
intdefault5Maximum number of clips Video Trim may return.
--max-seconds
intdefault45Maximum duration of each selected clip, in seconds.
--aspect
enumdefault9:16Output aspect ratio for every clip.
9:16 · 1:1 · 4:5 +1
9:161:14:516:9
--preset
enumdefaultserif-editorialCaption style used for every clip.
tiktok-bold-yellow · minimal-white · subtle-drop +3
tiktok-bold-yellowminimal-whitesubtle-dropkaraoke-gradientbig-serifserif-editorial
--corrections
stringdefaultOptional comma-separated word corrections in the form from=to,from=to.
--lang
stringdefaultautoISO 639-1 transcription language, or auto.
--parallel
intdefault4Maximum number of selected clips processed in parallel.
--position
enumdefaultautoCaption position. Auto keeps text away from the main subject when possible.
auto · top · middle +1
autotopmiddlebottom
How it works 5 pipelines
  1. transcription

    Creates the timestamped speech map Video Trim uses to keep every selected idea complete.

    Input source: OpenClaw Creator: Why 80% Of Apps Will Disappear
    Text 
    Today, I'm sitting down with Peter Steinberger, the creator of OpenClaw, the open source personal AI agent that has completely taken over the internet.
The GitHub repo exploded to over 160,000 stars practically overnight.
The community has built countless projects like Maltbook, where bots talk among themselves,
and now the bots are even renting humans to do tasks in the real world.
In our conversation, we discuss his aha moment, his contrarian development philosophies, and what this means for builders in 2026.
Let's dive in.
[upbeat music] So good to see you, man.
Hey, what's up?
Um, so you've made something people want.
…
  2. video-trim

    Reviews the complete source, selects the strongest moments from the editorial brief, and returns separate clips.

    Video
  3. video-reframe

    Reframes every selected clip to the chosen social aspect ratio while keeping speakers and important content visible.

    Video
  4. transcription

    Transcribes each reframed clip so its captions match the final edit and timing.

    Text 
    Are apps just gonna go away?
Uh, I think 80% of them are going away.
Why do I need MyFitnessPal?
Like, my agent already knows that I'm making bad decisions.
I'm at, I don't know,
uh, Smashburger something,
and it will already assume that I eat what I like to eat.
If I don't make a comment, it will just, like, automatically track it, or I make a picture and it'll just store it somewhere.
I don't even need to care, right?
  5. captions

    Burns the matching transcript into every clip with the selected style and position.

    Video
Why it works and what to change

Why it works

The recipe first creates a timestamped source transcript, then video-trim reviews the complete source, compares candidate moments against your instructions, and returns several self-contained clips. There is no separate highlight list to keep in sync and no timestamp file to prepare.

Every selected clip then gets its own finishing pass. Reframing happens before transcription, so the transcript and burned captions match the exact final cut. The branches run in parallel: asking for more clips adds work, but does not turn the recipe into one long serial queue.

The knobs that matter

  • instructions — describe both the subject and the editorial test. “Five strongest self-contained arguments” is better than “interesting moments.”
  • max_clips — caps the pack without forcing the agent to return weak filler when fewer moments satisfy the brief.
  • max_seconds — sets the ceiling for each clip. Give a shorter limit for hooks and a longer one for complete explanations.
  • aspect — use 9:16 for Shorts, Reels, and TikTok; switch to 1:1, 4:5, or 16:9 when the destination needs it.

The recipe options contain the caption style, language, corrections, and parallelism controls.

Notes

A strong source does not guarantee that every requested count exists. The agent may return fewer clips rather than repeat the same idea or cut a weak moment. Review each selection in context before publishing: an isolated quote can lose important qualifications from the original conversation.

The sample gallery was produced in one resumable run from the credited OpenClaw interview. It demonstrates the output format; AI selection can choose different moments on a new run.