Captions And Music Bed

Captions And Music Bed

Use this skill after the narration timeline is stable.

Captions

Caption source rules:

• Use the user's script as wording source.
• Use the final narration-only master as timing source.
• Preserve punctuation and capitalization.
• Show one caption line at a time.
• Start the caption when the first word is spoken.
• Avoid captions over silent hooks or silent CTA holds unless requested.

Preview style before burning the full video when style changes.

Read references/caption-rules.md for timing and style guidance.

Caption Style Defaults

• Bottom center.
• White text.
• Translucent dark backing.
• Generous bottom margin.
• Font size small enough not to dominate UI.
• One line only; split long sentences into natural phrase chunks.

Use templates/caption-style.ass as a starter.

Music Bed

Music source rules:

• Start from a chosen offset when the intro of the source track is weak.
• Keep hook music audible if there is no voiceover.
• Duck under narration.
• Do not raise music during later narration pauses if the user says it is distracting.
• End music at the requested time, including final silence if requested.
• If the music source ends early, restart from a musically useful tail section.

Read references/music-ducking.md for FFmpeg patterns.

QA

• Listen to start, middle, and final tail.
• Check music is present and audible, not merely technically in the stream.
• Confirm no clipped narration.
• Confirm captions are not late by several words.
• Extract a caption proof frame from a visually busy scene.
• Probe final audio stream.

Output

Return:

• Caption file path.
• Caption preview frame.
• Music bed strategy.
• Final mixed video path.
• Audio/caption checks performed.

Caption Chunking Rules

Split captions by:

• Natural phrase boundaries.
• Breath points.
• Punctuation.
• Maximum line length.
• Visual complexity behind the caption.

Do not split in a way that changes meaning. Do not add punctuation that the script does not contain.

Timing Sources

Preferred order:

1. Word-level alignment from final narration master.
2. Phrase timestamps from user or transcript.
3. Scene-level known timings plus manual phrase estimates.

Never time captions from an obsolete pre-edit audio file.

Caption Preview Procedure

Before full burn-in:

• Pick a visually busy frame.
• Render one still with caption style.
• Show or inspect it.
• Adjust font size, backing opacity, margin, and width.
• Only then burn the full video.

Music Bed Build Procedure

1. Probe source music duration.
2. Pick source start offset.
3. Decide hook volume.
4. Decide narration-bed volume.
5. Decide dense-explanation volume.
6. Decide whether pauses stay ducked.
7. Decide final CTA volume.
8. Decide music end timestamp.
9. Add fade out.
10. Mix with limiter.

Music QA

Listen/check:

• First 5 seconds.
• First narration entry.
• Dense explanation section.
• Long pause.
• Tail restart.
• Final 5 seconds.

If music is "technically there" but inaudible, raise bed or reduce ducking threshold.

Caption Failure Recovery

If captions are late:

• Check alignment source.
• Check whether silence/hook shifted global times.
• Start chunks at first word, not middle of phrase.
• Regenerate from final narration master.

If captions are too big:

• Reduce font size.
• Increase max characters slightly only if still one line.
• Lower backing opacity only if readable.

Handoff Format

Report:

• Caption source script.
• Caption timing source.
• ASS file path.
• Music source and offset.
• Ducking strategy.
• Final silence duration.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: saranambiar
• Source: saranambiar/hyperframes-video-agent-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.