Overview
This skill recuts a talking-head video — re-timing, cropping, or pacing segments — into a new composition via HyperFrames. Use it to tighten a long take into clips or repurpose footage for short-form. It renders the recut MP4 from the original source.
HyperFrames — Talking Head Recut
Layer timed, designed graphic cards onto an existing talking-head video — titles, lower-thirds, data callouts, quotes, side panels, PiP — synced to the transcript. The clip plays untouched underneath.
When to Use
- User wants graphic overlays on a talking-head video
- User wants lower-thirds with names/titles
- User wants data callouts synced to spoken numbers
- User wants pull-quotes or side panels
- User wants picture-in-picture layout
- User says "package my video" or "add graphics to my interview"
When NOT to Use
- User wants plain subtitles/captions (use
hyperframes-embedded-captions) - User wants to build a video from scratch (use
hyperframes-product-launch-video) - User wants a faceless explainer (use
hyperframes-faceless-explainer) - User wants to edit the footage itself (this skill leaves footage untouched)
Workflow
1 · Check environment
npx hyperframes doctor # ffmpeg, headless browser, render deps
ls "<SKILL_DIR>/assets/fonts" "<SKILL_DIR>/assets/vendor/gsap.min.js"
Required: ffmpeg / ffprobe (system), bundled fonts + GSAP (in skill assets).
2 · Create work directory
VIDEO_PATH="/absolute/path/input.mp4"
WORK_DIR="videos/$(basename "$VIDEO_PATH" | sed 's/\.[^.]*$//')"
mkdir -p "$WORK_DIR"
3 · Extract audio and metadata
ffprobe -v error -select_streams v:0 \
-show_entries stream=width,height,r_frame_rate \
-show_entries format=duration -of json "$VIDEO_PATH" > "$WORK_DIR/metadata.json"
ffmpeg -y -i "$VIDEO_PATH" -vn -acodec libmp3lame -q:a 2 "$WORK_DIR/audio.mp3"
4 · Transcribe (local Whisper)
npx hyperframes transcribe "$WORK_DIR/audio.mp3" -d "$WORK_DIR" --json --model small.en
Writes word-level transcript.json (word text + start / end timestamps). No API key needed.
5 · Correct transcript
Read transcript.json and fix obvious ASR errors (homophones, product names, technical terms). Preserve start / end timestamps.
6 · Draft storyboard
Design cards directly from the transcript. Each card has:
| Field | Purpose |
|---|---|
id |
stable id for HTML/GSAP selectors |
intent |
natural-language description |
startSec / endSec |
timing in seconds |
accentIndex |
which of 5 theme accent colors |
zone |
canvas position (fullscreen, lower-third, side-panel, whiteboard-area, video-overlay) |
contentHints |
kicker/title/detail/data/quote |
Card count formula:
secPerCard = basePace × densityMultiplier
cardCount = max(5, round(videoDurationSec / secPerCard))
| Video duration | Base pace |
|---|---|
| < 60s | 6-8s |
| 60s - 3 min | 8-12s |
| 3 - 10 min | 12-20s |
| 10 - 30 min | 20-35s |
| > 30 min | 30-60s |
| Density signal | Multiplier |
|---|---|
| High (many numbers, staccato) | × 0.7 |
| Medium (mixed) | × 1.0 |
| Low (one extended story) | × 1.5 |
7 · Decide render strategy
Ask the user to pick:
- Output ratio: 16:9 (wide source), 9:16 (tall source), 4:5 (near-square)
- Layout: standard, side-panel, picture-in-picture
- Style: noir, editorial, bold-poster, blue-professional, etc.
- Card density: auto (from formula), sparse, dense
8 · Design cards
Write each card's HTML directly in the conversation. Card zones:
| Zone | Bounds | Use for |
|---|---|---|
| fullscreen | whole canvas | hero moments, big numbers |
| lower-third | bottom 30% | annotation over visible video |
| side-panel | right 42% (landscape) / bottom 40% (portrait) | data side |
| whiteboard-area | inset 40px margin | dense data / annotated content |
| video-overlay | full canvas, mostly-transparent | annotation overlays |
9 · Assemble composition
Build index.html with:
- Video layer:
<video>element with the source footage - Card layers: absolutely-positioned divs with
data-start/data-duration - GSAP timeline: animating card entry/exit synced to transcript timestamps
<div id="video-wrap">
<video src="input.mp4" muted></video>
</div>
<div class="card" id="card-01" data-start="0.5" data-duration="12.5" data-track-index="1">
<div class="card-kicker">AN HONEST QUESTION</div>
<div class="card-title">The soul-searching question at 11 PM</div>
</div>
10 · Render
npx hyperframes render --output output.mp4
Verify the output:
ffmpeg -ss 10 -i output.mp4 -frames:v 1 frame-10s.png
Anti-Rationalization Table
| Rationalization | Reality |
|---|---|
| "I'll add a card every 2 seconds" | Use the formula — density matters, not arbitrary frequency |
| "I'll skip transcription and guess timings" | Transcript word timestamps are the clock — cards must sync to speech |
| "I'll use a fixed card structure" | Cards emerge from what the transcript actually says |
| "I'll edit the footage to match the cards" | Footage plays untouched — cards overlay, don't replace |
| "I'll center content at 42% to avoid overlap" | Cards are positioned in zones — video fills the frame |
| "I'll add an outro automatically" | No fixed brand outro — design one only if the user asks |
Verification
-
hyperframes doctorpasses -
transcript.jsonhas word-level timestamps - Card count ≥ 5 (floor)
- Each card has
startSec<endSec - Card timings clamped to video duration
- Output MP4 has visible synced overlays on 3+ sampled frames
- Footage plays untouched underneath