jalyn-vertical-talk-video-edit

name	jalyn-vertical-talk-video-edit
description	Use this skill when editing a 9:16 vertical Chinese talking-head video with Codex. It guides short-video opening hooks, face-safe layouts, large subtitles, phrase-triggered cards, material cut-ins, speaker PIP, vertical Remotion rendering, and mobile-first quality checks.

Jalyn Vertical Talk Video Edit

Goal

Turn a vertical talking-head video into a clean short-form explainer. The first screen must make the topic obvious, subtitles must stay readable on mobile, and all components must protect the face, mouth, microphone, and CTA area.

This skill is a vertical editing standard. It can guide a Remotion render if a template exists. Without a template, it should still produce a scene plan, caption timing, material placement plan, and quality checklist.

Required Inputs

• Source talking-head video.
• Original script, if available.
• Material folder, if assets are provided.
• Topic title and CTA.
• Output target: 1080x1920, 30fps, H.264/AAC unless the user asks otherwise.

Workflow

1. Inspect source metadata with ffprobe.
2. Extract a contact sheet to judge face position, headroom, hand gestures, and subtitle safe zone.
3. Transcode unstable inputs to vertical H.264/AAC if needed.
4. Transcribe audio and correct the transcript with the original script.
5. Segment the script into short semantic beats.
6. Decide whether each beat uses:
   • full speaker
   • content canvas + speaker PIP
   • statement card
   • material cut-in
   • CTA card
7. Keep the first 1-2 seconds extremely clear: title, problem, or result.
8. Assign fixed slots for title, caption, chips, material, and PIP.
9. Trigger cards and material by phrase, not by average chapter duration.
10. Render stills on mobile-safe frames.
11. Render full video and generate a vertical contact sheet.

Vertical Layout Rules

• Top area: short topic title, progress, or chapter label only.
• Middle area: speaker face and core visual; never block the face.
• Lower-middle area: temporary cards or material cut-ins, only if they do not block mouth and subtitles.
• Bottom area: subtitles and CTA; keep it reserved.
• PIP should be above the subtitle area, usually bottom-right or bottom-left.

Scene Decisions

• Use full speaker when the person is making a key point, hook, identity statement, or CTA.
• Use content canvas + PIP when screenshots, prompts, covers, or process visuals are the main subject.
• Use statement card for a strong conclusion or contrast sentence.
• Use material cut-ins only when the material is readable after cropping.
• Do not force wide screenshots into vertical full screen. Crop the important area or remake it as a vertical card.

Animation Rules

• Keep motion short and phrase-synced.
• Card reveal: 6-9 frames.
• Component lead: 0.10-0.16s.
• Do not stack multiple animated elements on the same word.
• One beat gets one main animation.
• If a material appears, the previous material must exit, shrink, or move away first.
• Do not use large flying cards that cross the face.
• Do not put labels or decorative text inside the speaker PIP.

Subtitle Rules

• Subtitles are the primary reading layer.
• Prefer 1-2 lines.
• Avoid three-line captions unless absolutely necessary.
• Highlight only 1-2 keywords.
• Check the final 10 seconds manually, especially CTA words and numbers.

Material Rules

• Screenshots must be cropped to mobile-readable areas.
• Covers can be shown as result cards, not necessarily full screen.
• Prompt screenshots should be cropped to the command and key steps.
• Skill docs should be used as proof near the end, with highlighted file names or titles.
• If a material cannot be read on a phone, replace it with a simplified card.

Quality Gates

Fail the edit if any of these happen:

• The first screen does not reveal the topic.
• Subtitles cover the mouth.
• Cards or stickers block eyes, mouth, glasses, microphone, or hands.
• PIP crops the face badly.
• Screenshots are unreadable on mobile.
• Several cards overlap and make the frame feel cluttered.
• CTA caption does not match the voice.

Validation Commands

Use equivalents when paths differ.

npx tsc --noEmit

npx remotion still src/index.ts <CompositionId> out/check-vertical.png --frame=<frame>

npx remotion render src/index.ts <CompositionId> out/output-vertical.mp4 --crf=18 --concurrency=1

ffmpeg -y -i "out/output-vertical.mp4" -vf "fps=1/5,scale=216:-1,tile=5x4:margin=6:padding=4:color=white" -frames:v 1 "out/contact-vertical.jpg"

Delivery Note

When reporting completion, include:

• Output path.
• 1080x1920, fps, codec, duration.
• Contact sheet path.
• Whether the face, subtitle area, and CTA were checked.
• Any source-quality limitations.