AI Video Clipper: YouTube to Shorts, Reels & TikTok + Captions

Turn one long video into finished vertical Shorts automatically. Paste a YouTube link to a podcast, interview, keynote or stream. The AI Video Clipper finds the strongest moments, cuts each one into a 9:16 clip that makes sense on its own, tracks the speaker's face, burns in word-by-word captions and a hook title, and writes an SEO-ready title, description and tags for every clip.

$0.99 per Short delivered, with no subscription, no editor and no login. If no Short is delivered, nothing is charged.

What you get for each Short

Video	MP4, 1080×1920 (9:16), H.264 + AAC, loudness-normalized to -14 LUFS (YouTube's playback target)
Edit	Cold open: the most striking sentence plays first, then the context, then the payoff. Filler and tangents are cut.
Framing	Face-tracked vertical crop that re-frames on every camera change. Branded borders and letterboxing are trimmed automatically.
Captions	Word-by-word, active word highlighted, big bold type built for phone screens
Hook title	On screen for the first 3.6 seconds, placed below the face
Credit	Speaker and show name on screen, plus a link to the full source video in the description
Metadata	A title that names the speaker, a 2-3 sentence description stating who said what, and 12-18 search-phrase tags
Why this moment	One line on why a stranger would stop scrolling for it

How it works

1. Transcribe. The whole video is transcribed with Whisper (large-v3), with a timestamp on every word.
2. Sentence-align. The transcript is rebuilt into real sentences, so no cut ever starts or ends mid-sentence.
3. Pick. An AI editor reads the full transcript and proposes the strongest self-contained moments: surprising claims, stories and quotable lines that need no context.
4. Quality gate. Code rejects any proposal that is too short or too long (18-62 s), doesn't end on a full sentence, lacks metadata, or credits a person who appears nowhere in the video's title or transcript. The AI can't invent a speaker.
5. Render. Only the chosen seconds are used. Each Short is rendered frame by frame at 1080p with face tracking, captions and the hook.

How long does it take, and why?

Measured on Apify's servers:

Source video	Shorts	Run time
3.4-minute interview	1	about 3.3 minutes
66-minute podcast	2	about 10 minutes

Most of the time goes to three things:

• Downloading and transcribing the whole video. The AI has to hear every minute to find the best one, so a longer source takes longer.
• Reading the full transcript to choose moments.
• Rendering every frame at full 1080×1920 with face tracking and captions: about 1-2 minutes per Short.

Asking for more Shorts from the same video adds only render time, because the video is downloaded and transcribed once. Start the run and come back to finished files. The default run timeout is 60 minutes; raise it for sources over about 2 hours.

Input

Field	Default	What it does
videoUrl	a 3-minute public-domain sample	A YouTube link, or any page or file yt-dlp can read (Vimeo, X/Twitter, direct .mp4)
maxShorts	1	Up to this many Shorts (1-10). The gate may return fewer if the video has fewer strong moments. You pay only for Shorts delivered.
speakerName	blank	Who to credit. Blank lets the AI name the guest, checked against the title and transcript.
showName	channel name	Shown in the on-screen credit

Copy-paste examples

Input JSON. Three Shorts from a podcast, crediting the guest:

{
  "videoUrl": "https://www.youtube.com/watch?v=kwSVtQ7dziU",
  "maxShorts": 3,
  "speakerName": "Andrej Karpathy",
  "showName": "No Priors"
}

curl. Start a run and wait for the results in one call (replace YOUR_TOKEN):

curl -X POST "https://api.apify.com/v2/acts/EliAI~longform-to-shorts/run-sync-get-dataset-items?token=YOUR_TOKEN&timeout=3600" \
  -H "Content-Type: application/json" \
  -d '{"videoUrl": "https://www.youtube.com/watch?v=kwSVtQ7dziU", "maxShorts": 2}'

JavaScript / Node.js (npm install apify-client):

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('EliAI/longform-to-shorts').call({
  videoUrl: 'https://www.youtube.com/watch?v=kwSVtQ7dziU',
  maxShorts: 2,
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
for (const s of items.filter((i) => i.ok)) {
  console.log(s.title, s.videoUrl);   // MP4 link + ready-to-post title
}

Python (pip install apify-client):

from apify_client import ApifyClient

client = ApifyClient("YOUR_TOKEN")
run = client.actor("EliAI/longform-to-shorts").call(run_input={
    "videoUrl": "https://www.youtube.com/watch?v=kwSVtQ7dziU",
    "maxShorts": 2,
})
for s in client.dataset(run["defaultDatasetId"]).iterate_items():
    if s["ok"]:
        print(s["title"], s["videoUrl"], s["tags"])

Download every MP4 from a finished run (bash + jq):

curl -s "https://api.apify.com/v2/datasets/DATASET_ID/items?clean=true" \
  | jq -r '.[] | select(.ok) | .videoUrl' | xargs -n1 curl -sO

n8n / Make / Zapier. Use the Apify app's "Run Actor" action with EliAI/longform-to-shorts and the input JSON above, then map videoUrl, title, description and tags straight into a YouTube, TikTok or Instagram upload step. Point a scheduled trigger at a channel's newest video to get Shorts every time a new episode drops.

Output

One dataset row per Short. The MP4 is in the run's key-value store, linked from videoUrl.

{
  "ok": true,
  "shortNumber": 1,
  "videoUrl": "https://api.apify.com/v2/key-value-stores/<store>/records/short-1-jensen-huang-culture-is-a-choice.mp4",
  "title": "Jensen Huang: Culture is a Choice",
  "description": "Jensen Huang, CEO of NVIDIA, explains that company culture is a direct result of the choices made during the employee selection process. He details the company's intense selectivity in hiring, offering to only 10% of interviewees with a 95% acceptance rate...",
  "tags": ["jensen huang", "nvidia", "company culture", "employee selection", "hiring process", "leadership", "..."],
  "durationSeconds": 50.88,
  "whyThisMoment": "A CEO reveals that company culture is a deliberate choice, not an accident.",
  "sourceUrl": "https://www.youtube.com/watch?v=Pnr_TVjYlj4",
  "sourceRangesSeconds": [[69.58, 75.0], [76.02, 99.235], [99.235, 121.435]],
  "width": 1080,
  "height": 1920
}

Pricing

• $0.99 per Short delivered. This is the main charge.
• $0.25 per extra hour of source video past the first, charged only when at least one Short is delivered. A 66-minute podcast adds $0.25; a 20-minute interview adds nothing. It covers the download and transcription time that long sources need.
• Never charged: runs that deliver no Short, moments the quality gate drops, videos that can't be downloaded.

Example: 3 Shorts from a 90-minute podcast = 3 × $0.99 + 1 × $0.25 = $3.22.

Use cases

• Podcasters and YouTubers: turn every episode into a week of Shorts, Reels and TikToks.
• Clipping agencies and clipping campaigns: batch long videos into ready-to-post clips via the API.
• Social media managers: repurpose webinars, keynotes and interviews without an editor.
• Automation builders: call it from n8n, Make, Zapier or your own code, then post with the YouTube Data API using the returned title, description and tags.

FAQ

How do I turn a YouTube video into Shorts automatically? Paste the video link into videoUrl and run. The AI Video Clipper transcribes the video, picks the best moments, and returns vertical 1080×1920 MP4s with captions, plus a title, description and tags for each.

How do I turn a podcast into YouTube Shorts, Reels and TikToks? Use the podcast's YouTube (or video file) link. Podcasts and interviews are what it's built for: someone speaking on camera, with clear moments to cut.

Is this an OpusClip alternative? It does the same job (long video in, captioned vertical clips out), but you pay per Short with no subscription, and it runs as an API you can automate.

Does it add captions / subtitles? Yes. Word-by-word captions are burned into every Short, with the spoken word highlighted as it's said.

Can I schedule the Shorts to YouTube? The output gives title, description and tags in the format YouTube's upload API expects. Upload as private with a publishAt date to schedule.

What videos work best? Podcasts, interviews, panels, keynotes, lectures, streams and commentary. Music videos, silent footage and slides without a visible speaker are not a good fit.

What languages? English works best. Other languages transcribe, but moment-picking quality varies.

Why did my run return fewer Shorts than I asked for? The quality gate drops moments that cut mid-sentence or run too short or long. You are only charged for Shorts delivered.

Limits and fair use

• Sources up to about 3 hours (raise the run timeout for long sources).
• Clip only content you have the right to use. Every Short credits the speaker and show on screen and links the full source in its description.

Changelog

• 0.2 (2026-09-27): first public release. Sentence-aligned cuts, face-tracked crop with automatic border trimming, word-by-word captions, hook title, AI-named speaker checked against the title and transcript, SEO/GEO metadata per Short.