AI Clipping

One API call: long video in → ranked vertical short clips out.

Each clip ships with a viral score (0–100), an opening hook line, a one-sentence "why it works" reason, and a hosted mp4 URL.

Underlying API: https://muapi.ai/playground/ai-clipping Reference implementation (open source): https://github.com/SamurAIGPT/AI-Youtube-Shorts-Generator

When to Use

• Auto-clip a podcast, interview, lecture, vlog, or stream into TikTok / Reels / Shorts.
• Extract the best 30–75s moments from any hosted video URL.
• Get face-tracked vertical (9:16), square (1:1), or portrait (4:5) crops without running ffmpeg locally.

If you only need raw timestamps for your own renderer, set --coords-only to skip cropping and just get the highlight ranges.

Agent Execution Protocol

Step 1 — Collect Inputs

If the user gave only a video URL, run with defaults — don't block on questions.

Step 2 — Verify Prerequisites

• muapi-cli installed and authed (muapi auth configure)
• MUAPI_API_KEY available (env var or muapi auth status passes)

That's it. No ffmpeg, no Python, no Whisper install, no LLM keys. Everything runs server-side.

Step 3 — Run the Skill

bash library/edit/ai-clipping/scripts/run-ai-clipping.sh \
  --video "https://example.com/podcast.mp4" \
  --num-clips 5 \
  --aspect-ratio 9:16 \
  --view

The script:

1. Resolves --video to a hosted URL (uploads local files via muapi upload file if needed).
2. Calls muapi edit clipping with the supported parameters.
3. Polls until the job is done (or returns the request_id immediately under --async).
4. Prints a ranked summary and, if --output-json is set, writes the full result.

What Happens Server-Side

The /ai-clipping endpoint internally runs the full pipeline so the agent doesn't have to:

• Transcribe with Whisper.
• Classify content type (podcast / interview / tutorial / vlog / lecture / monologue).
• Rank highlights through the virality framework:
• Hook moments — strong opening line that stops the scroll
• Emotional peaks — laughter, anger, vulnerability, awe
• Opinion bombs — spicy, contrarian, debate-bait takes
• Revelation moments — "wait, what?" reframes
• Conflict — disagreement, tension, callouts
• Quotable lines — tight, screenshot-worthy phrasing
• Story peaks — climax of a narrative arc
• Practical value — actionable insight a viewer will save
• Dedupe overlapping candidates by score.
• Top-N select and face-track auto-crop to the requested aspect ratio.

This is why the skill is small: the heavy lifting is on the API.

Quick Invocation Patterns

Defaults — three 9:16 clips:

bash run-ai-clipping.sh --video "https://example.com/long.mp4"

Podcast — more clips, view in player:

bash run-ai-clipping.sh --video "<URL>" --num-clips 8 --view

Square clips for Instagram feed:

bash run-ai-clipping.sh --video "<URL>" --aspect-ratio 1:1 --num-clips 3

Just the timestamps (build your own renderer):

bash run-ai-clipping.sh --video "<URL>" --coords-only --output-json result.json

Async submit (returns request_id, poll later):

REQUEST_ID=$(bash run-ai-clipping.sh --video "<URL>" --async --output-json - | jq -r '.request_id')
muapi predict wait "$REQUEST_ID" --download ./outputs

Local file:

bash run-ai-clipping.sh --video ./recording.mp4 --num-clips 5 --view

Batch — `urls.txt` with one URL per line:

xargs -a urls.txt -I{} bash run-ai-clipping.sh --video "{}"

Aspect Ratio Picker

Default to 9:16 unless the platform is specified.

Output Schema

{
  "source_video_url": "...",
  "shorts": [
    {
      "title": "The one mistake that cost me $50K",
      "start_time": 124.3,
      "end_time": 187.6,
      "score": 92,
      "hook_sentence": "Nobody talks about this, but it killed my first startup...",
      "virality_reason": "Opens with a number + regret, peaks on a contrarian lesson",
      "clip_url": "https://.../short_1.mp4"
    }
  ]
}

When --coords-only is set, each entry has start_time/end_time but no clip_url — render locally with ffmpeg.

When reporting back to the user, surface for each clip: rank, score, time range, title, hook, and clip URL.

Common Mistakes to Avoid

1. Wrong aspect ratio for the platform — Shorts / TikTok / Reels are 9:16. Default to that.
2. Padding to hit `num_clips` — if the API returns fewer survivors than requested, return what you have. Don't pretend.
3. Re-running on a 404'd clip URL — the same request_id can be re-fetched with muapi predict wait <id> rather than re-clipping.
4. Trying to tune Whisper / chunk size / LLM prompts — those knobs aren't exposed; the endpoint handles them.

Failure Modes

• API key missing or rejected — surface the exact error; never fabricate a key.
• Job timed out — bump poll timeout (--poll-timeout) and retry.
• Source URL not reachable from the backend — upload locally with muapi upload file <path> first, then pass the returned URL.
• Fewer clips returned than requested — the source had fewer rankable highlights. Return what came back with a note.

Done Criteria

The skill is done when:

1. result.shorts has up to num_clips entries, each with a working clip_url (or start_time/end_time under --coords-only).
2. The user has been shown the ranked list (score, time range, title, hook, URL).
3. If --output-json was set, the file exists and parses.