Short-form Repurposing

Mission

Turn an authorized long-form source into a small, ranked set of coherent short-form candidates. Preserve the source meaning, identify useful start/end points, adapt framing, and caption only after the candidate segment is known. This is editorial packaging, not automatic publication.

Required inputs

• source video URL or local file and provenance/rights state;
• platform or target ratio, language, caption style, and delivery format;
• desired clip count, approximate length, audience, hook, and CTA;
• whether source audio and wording must remain verbatim; and
• review owner, budget, and definition of done.

If no platform is named, return a neutral 9:16 candidate plus the assumption; do not claim that a crop is platform-compliant without a supplied rule set.

Pipeline

1. Inspect or upload the source. Record duration, dimensions, audio, language, and source ID. Never overwrite the original.
2. Use media.clip_video/AI clipping when available to propose highlights. Capture each candidate's timestamps, transcript or rationale, score if supplied, and source parent.
3. Rank for standalone meaning, hook clarity, payoff, audio quality, visual legibility, and overlap with other candidates. Deduplicate overlapping moments. Return fewer clips rather than padding the requested count with weak or misleading moments.
4. Choose exact start/end times and use media.reframe_video or the live autocrop schema for the requested ratio. Check faces, products, text, and speaker direction after the crop.
5. Generate captions with media.caption_video only after selecting the segment. Preserve supplied wording; flag uncertain transcription, names, numbers, and claims for human review.
6. Review the resulting clip for context loss, jump cuts, crop safety, caption timing, audio clipping, duration, and export. Use video QA.

The current MuAPI utility schemas differ: AI clipping accepts a source video and highlight count, autocrop accepts explicit start/end times and a ratio, and captions accept provider/language/theme fields. Resolve them through MODELS.md and muapi-video-tools.md; do not assume a single tool performs all three stages.

Output contract

Return a ranked candidate table with clip ID, source ID, exact timestamps, reason/hook, transcript status, crop ratio, caption status, model/tool, prediction IDs, output URLs, QA notes, and rejected/failed candidates. Include the source link and parent/child relationships. Save an asset manifest and receipt under .video/ only when permitted, following reports-and-receipts.md.

Never change the source's meaning, remove qualifying context, or imply that captions or crops are final without review. Publishing remains out of scope.