Caption Studio
Word-by-word captions that carry a talking-head clip.
Beat-synced lyric videos for the vertical feed.
Builds a lyric video that actually lands on the beat. Upload a track and its lyrics, and the skill detects tempo and sections, times each line, and animates type that reacts to the music. Vertical is the default, with a landscape version for YouTube and artist pages.
Preview format
Beat Lyrics
Rendered by this skill · unedited
Placeholder clips, not real skill output. The 9:16 view shows approximate platform safe zones — always check the final render in the app.
Example commands for this skill. Copy one into your agent.
Use beat-lyrics with track.mp3 and lyrics.txt, make a 9:16 teaser from the chorus, and export a 16:9 full version. Use beat-lyrics to make a 5-second 9:16 test run. Use beat-lyrics to render the 16:9 master from the same project, with no copy changes. Everything the skill produces from a single run.
An audio file and lyrics text, plus optional artist artwork and color direction.
Know before you run it.
Run the doctor command after install to verify the environment.
| Tool | Version | Status | Purpose |
|---|---|---|---|
node | >=20 | Required | Runs the skill's build scripts and render pipeline. |
ffmpeg | any | Required | Analyzes audio and encodes the final video. |
chrome | any | Required | Headless Chrome renders the lyric composition. |
FAL_KEY | any | Optional | Optional generated background visuals via FAL. |
Rendering is local and free. Optional FAL jobs are billed by FAL per job; the skill lists costs and waits for approval.
The skill shows an estimate and waits for approval before any paid job starts.
We only mark an agent as tested after a clean install and render.
| Agent | Status | Install directory | Model |
|---|---|---|---|
| Claude Code | Tested | ~/.claude/skills | Claude Sonnet 4.5 |
| Cursor | Tested | ~/.cursor/skills | Claude Sonnet 4.5 |
| Codex | Tested | ~/.agents/skills | Claude Sonnet 4.5 |
| Gemini CLI | Untested | ~/.gemini/skills | — |
| OpenCode | Untested | ~/.agents/skills | — |
Run the same prompt to reproduce this preview. Output is rendered by the skill, unedited.
Use beat-lyrics with track.mp3 and lyrics.txt, make a 9:16 teaser from the chorus, and export a 16:9 full version. What the agent does, step by step.
Analyze tempo, beats and song sections from the audio.
Align lyrics to lines and words, then review ambiguous timings.
Choose a type treatment that matches the track's energy.
Animate line changes, accents and background reactions on the beat.
Render the 9:16 master and a 16:9 version.
Check readability, safe zones and audio sync.
Each skill renders the formats its target platforms expect.
Default 9:16
Full-screen vertical with safe zones.
Vertical with caption-safe framing.
Vertical with top and bottom clearance.
Square feed and carousel covers.
Landscape master and thumbnails.
One entry, derived from the version and the last tested date.
Verified the install prompt, the 9:16 / 16:9 exports and the safe-zone check on Claude Code, Cursor and Codex. Last run on 2026-09-12.
Same scenarios and output, different angle.
Word-by-word captions that carry a talking-head clip.
A product URL in, a 60–90 second launch film out.
CSV in, animated chart story out.