| name | X Video to Vertical Social Cut |
| description | Turn an X/Twitter video or local source clip into a verified 9:16 social export: preserve widescreen foreground footage, use a dark blurred background, burn in phone-readable captions, and omit persistent headers by default. |
X Video to Vertical Social Cut
Use this workflow when Raul supplies an X/Twitter video URL or a local video and wants a vertical social cut with captions.
Non-negotiable execution lane
Use the proven direct FFmpeg/run-command media workflow for all source footage. Download/resolve the real MP4, inspect it, select the source time range, generate the blurred background and burned-in subtitles with direct FFmpeg filters, export H.264/AAC, and QA the actual MP4.
- Do not use HyperFrames, HTML Motion, Remotion, or frame-by-frame HTML rendering to carry, crop, caption, or render the source video.
- Do not open a Creative/HyperFrames composition for this task unless Raul explicitly asks for an additional designed overlay, title card, or motion-graphic layer. Even then, composite that layer over the already-rendered direct-FFmpeg source cut rather than routing the underlying footage through it.
- This workflow’s known-good path is the direct command/media path. Prefer it over generic Creative video tools for speed and fidelity.
Output contract
- Produce an H.264/AAC 9:16 MP4, normally 1080×1920 at 30fps unless source or delivery constraints require otherwise.
- Preserve the full widescreen source video as the centered foreground. Do not crop away meaningful UI or cinematic action.
- Fill unused vertical space with a darkened, blurred version of the same footage.
- Burn in large, high-contrast captions in the lower safe area. Keep captions clear of mobile UI and never clip them.
- No persistent header, title strip, or watermark by default. Add one only when the user explicitly asks.
- Use
faststart / mobile-ready MP4 settings when export tooling supports it.
Workflow
- Resolve/download the video. For X status URLs, use the X/media recovery path; if regular post extraction fails, use the media downloader/resolver fallback rather than abandoning the source.
- Inspect the source video and transcript. Prefer Creative transcription; cloud STT is primary, and local Whisper is the recovery path when cloud transcription is unavailable.
- Choose a coherent 20–45 second moment with a clear hook and enough dialogue/context. Use the user’s requested length when specified.
- Build the vertical composition: blurred/dark background, centered uncropped foreground, captions in lower safe area, no default header.
- Export H.264/AAC.
- QA before delivery:
- technical probe/full decode succeeds;
- sample early, middle, and ending frames or use a contact sheet;
- visually confirm no black frames, accidental header, crop loss, or caption clipping;
- verify aspect ratio, duration, audio track, and caption readability;
- if the generic scorer is confused by source-video metadata, do not use it as the sole acceptance gate.
- Deliver the final MP4 and report only verified facts. Mention any genuine uncertainty, especially if audio can be transcribed but not directly auditioned.
Proven Odyssey reference
The verified no-header Odyssey test used full cinematic widescreen footage centered over a dark blurred vertical background, bold outlined all-caps captions in the lower safe area, 1080×1920 H.264/AAC, 30 seconds. Visual contact-sheet review found no persistent header, black frames, crop loss, or caption clipping. Local Whisper fallback produced the transcript during QA.
Avoid
- Do not leave a persistent title/header on screen unless explicitly requested.
- Do not crop the foreground aggressively just to fill 9:16.
- Do not claim visual or audio QA without actual evidence.
- Do not give up on an X URL solely because the initial post fetch lacks media metadata.
