| name | jalyn-horizontal-talk-video-edit |
| description | Use this skill when editing a 16:9 horizontal Chinese talking-head video with Codex. It guides source inspection, script-aligned captions, scene segmentation, full-screen speaker moments, large material boards, picture-in-picture speaker anchors, phrase-triggered animation, Remotion rendering, and pre-delivery quality checks. |
Jalyn Horizontal Talk Video Edit
Goal
Turn a horizontal speaker video into a polished explainer or tutorial video. The speaker is the trust anchor; screenshots, covers, product UI, prompts, and method boards are supporting evidence.
This skill is an editing standard, not a standalone renderer. If a Remotion project is available, use it for rendering. If not, produce a clear edit plan, timestamped captions, scene map, asset layout, and quality checklist.
Required Inputs
- Source talking-head video.
- Original script, if available.
- Material folder, if assets are provided.
- Desired CTA.
- Output target:
1920x1080, 30fps, H.264/AAC unless the user asks otherwise.
Workflow
- Inspect the source with
ffprobe.
- Extract frames near
1s, 8s, and the final section to judge face position and safe zones.
- Transcode HEVC or unstable phone footage to
1920x1080 H.264/AAC before using Remotion.
- Extract audio and transcribe with Whisper or faster-whisper.
- Correct the transcript with the user's original script. Pay special attention to names,
Codex, Skill, product names, numbers, and CTA words.
- Segment the script by meaning, not by equal duration.
- Choose a layout for each segment:
- full speaker
- large material board + speaker PIP
- full-screen method board
- result/proof board
- CTA
- Assign every material to a fixed slot before animating it.
- Create props or an edit timeline with caption times, scene times, and phrase-triggered items.
- Render stills for key sections before the full render.
- Render the full video.
- Generate a contact sheet and check face safety, overlap, caption timing, and asset readability.
Layout Rules
- Use full speaker for opening trust, major turns, identity statements, and CTA.
- Use large material board + speaker PIP when screenshots, prompt images, product UI, or Skill docs need to be read.
- Use full-screen method board for frameworks, steps, comparisons, and conclusions.
- Do not show a screenshot if the viewer cannot read its core content.
- Crop and zoom screenshots instead of forcing the full screenshot into the frame.
- Keep subtitles in the bottom safe area.
- Do not place cards over eyes, mouth, microphone, or the active hand gesture zone.
Animation Rules
- Caption lead:
0.03-0.06s.
- Component lead:
0.10-0.18s.
- Card reveal:
7-10 frames.
- Scene transition: under
18 frames.
- One semantic beat gets one main animation only.
- Avoid simultaneous movement of prompt card, screenshot, toast, PIP, and caption.
- Before a new large asset enters, fade, shrink, or move the previous asset to a secondary slot.
- Never let new assets directly cover readable text from the previous asset.
Material Slots
Use stable slots:
- Main visual slot: large screenshot, product page, cover, or process board.
- Secondary visual slot: smaller result card, comparison card, or supporting asset.
- Speaker PIP slot: clean circular or rounded portrait, no labels over the face.
- Caption slot: bottom center.
- CTA slot: final action card only.
If two large materials compete, pick the one that matches the current sentence and delay the other.
Quality Gates
Fail the edit if any of these happen:
- Captions lag or lead noticeably.
- Face, mouth, or eyes are blocked.
- Screenshots are too small to read.
- Materials overlap in a way that hides important content.
- Template words such as
Hook, Action, Scene, version labels, or debug labels appear.
- The final CTA text does not match the voice.
Validation Commands
Use equivalents when paths differ.
npx tsc --noEmit
npx remotion still src/index.ts <CompositionId> out/check.png --frame=<frame>
npx remotion render src/index.ts <CompositionId> out/output.mp4 --crf=18 --concurrency=1
ffmpeg -y -i "out/output.mp4" -vf "fps=1/5,scale=480:-1,tile=4x3" -frames:v 1 -update 1 "out/contact.jpg"
Delivery Note
When reporting completion, include:
- Output path.
- Resolution, fps, codec, duration.
- Whether still checks and contact sheet checks were done.
- Any remaining risk, especially source-video clarity or unreadable provided screenshots.
