lyric-video
Lyric-video editor: times every lyric line to the sung vocal (Whisper word timestamps, optionally on a Demucs vocal stem), shows each line with a karaoke fill as it is sung and the next line dimmed, over artwork, footage or a beat-reactive background, and delivers a 16:9 lyric vi
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/lyric-video
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
git clone https://github.com/Viprasol-Tech/claude-video-editing-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole viprasol-tech/claude-video-editing-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Lyric Video
The words land exactly when they are sung, readable on any background, in the artist's look.
- Best for: song releases, sing-alongs, devotional and worship lyrics, karaoke versions
- Inputs: the song (master), the approved lyrics text, cover art / background footage
- Default skin:
neon-orbit(use--no-matte) · alternativespaper-cut,swiss-poster(--no-matte),minimal-mono - Formats: 16:9 1920x1080, 9:16 chorus clip (15-30 s), lyrics
.srt - Duration target: the song; genre preset
lyric-video
Before you start
- Needs video-studio-core in the same skills folder.
CORE=video-studio-core,SKILLS= the folder holding both. - First time:
python CORE/scripts/check_env.py. Preset:python CORE/scripts/editor.py genre lyric-video. - Get the lyrics as text, exactly as the artist writes them (spelling, line breaks, censoring). Whisper only supplies timing; the words on screen come from the approved text.
- Vocal stem (optional, much better timing on dense mixes):
pip install demucs.
How a pro cuts this
| section | what happens |
|---|---|
| intro | title + artist on the artwork until the first sung word |
| verses | one line at a time, low-to-mid frame, calm motion |
| chorus | bigger type or a different layout; the same design each chorus so it feels like a hook |
| bridge | a visual change (colour, background) |
| outro | title card again, credits |
- Timing: each line appears 0.1-0.2 s before its first sung word and leaves 0.2 s after the last; words fill as sung. No line on screen < 1 s.
- Line length: <= 42 characters (16:9), <= 26 (9:16); split long lines at the singer's breath.
- Readability: contrast over any background (shadow/box/dim the background with
vignette); avoid the face if footage has a singer. - Motion: gentle (
maskUp,fade); beat pulses on the chorus only; never shake text people must read. - Audio: the song untouched (
"audio": {"voice": false}), normalised to -14 LUFS. - Colour: follow the cover art; a look on footage only (
music-video-neon,dreamy,vintage-70s). - Client (artist) judges: sync, readability, matches the release artwork.
- Common mistakes: lyrics typed from Whisper (wrong words), lines late by half a second, two lines changing at once, tiny type, text over the singer's mouth.
Make the edit
- A background video carrying the song (artwork still; or footage cut to the song with
timeline.py/ themusic-videoskill):
ffmpeg -loop 1 -i cover.jpg -i song.wav -r 30 -pix_fmt yuv420p -c:v libx264 -tune stillimage -c:a aac -shortest lyr_src.mp4
- Plan with word timing from the song:
python CORE/scripts/run.py lyr_src.mp4 --style SKILLS/neon-orbit --genre lyric-video --no-cut --no-matte --format 16:9 \
--model small --prompt "distinct words from the lyrics" --work lyr --stop-after plan
- If the timing is poor (busy mix), time the vocal stem instead, in its own folder, and use its words for timing:
python CORE/scripts/stems.py song.wav --out lyr_vox/stems
ffmpeg -loop 1 -i cover.jpg -i lyr_vox/stems/vocals.wav -r 30 -pix_fmt yuv420p -c:v libx264 -c:a aac -shortest lyr_vox/vox.mp4
python CORE/scripts/prep.py lyr_vox/vox.mp4 --work lyr_vox && python CORE/scripts/transcribe.py --work lyr_vox --model small
# read lyr_vox/transcript.json for line start/end times (song time = video time, both start at 0)
- Write the
lyricelement from the approved text + those times (next section), then:
python CORE/scripts/render.py lyr/plan.json --frames 10,30,60,90
python CORE/scripts/render.py lyr/plan.json
python CORE/scripts/finish.py lyr/plan.json --platform youtube
Always pass --format matching the project: run.py and make_plan.py otherwise use the skin's own format (many skins default to 9:16).
Refine the plan
- Remove the skin's
captionselement. - Lyric element:
{"type": "lyric", "start": 0, "end": 212.0, "y": 760, "w": 1500, "lines": [{"text": "Seek it anywhere", "start": 31.6, "end": 33.0}, {"text": "you will find me there", "start": 33.2, "end": 35.4}]}- text from the approved lyrics, times from the transcript (start 0.1 s early). - Per-word fill: give a line its own
words([{"w": "Seek", "start": 31.6, "end": 31.9}, ...]) copied from the transcript but spelled as in the lyrics. - Title / credits:
titlecardwith artist and song at 0 s until the first line; credits at the end. - Chorus emphasis: a second
lyricelement for chorus ranges with a biggerstyle({"fontSize": "96px"}) and"pulseOn": "beats"(add"beats"frombeats.py song.wav --out lyr/beats.jsonviamake_plan.py --beats). - Background:
vignette("strength": 0.7) for contrast;lightleak("intensity": 0.3) in the bridge. - Audio:
"audio": {"voice": false}; nomusicblock (the song is the source audio). - SRT of lyrics: put the approved words with their times into
plan.wordssodeliver.pywritescaptions.srtfrom them.
Deliver
python CORE/scripts/deliver.py lyr/plan.json --to youtube --burn no --out-dir lyr/deliver --no-post
python CORE/scripts/qc.py lyr/deliver/final_youtube.mp4 --plan lyr/plan.json --platform youtube
- 9:16 chorus clip: a new work folder with
--format 9:16on the chorus range (prep.py lyr_src.mp4 --work chorus --start 58 --end 84, thenrun.py chorus/source.mp4 ... --format 9:16), lines <= 26 characters. - Give the artist
captions.srt(also usable as synced lyrics on platforms that accept SRT).
Check before delivering
- Every word matches the approved lyrics (diff your
linesagainst the lyric sheet). - Sync checked at the first line, each chorus start and the last line (+-0.1 s).
- Readable on the darkest and brightest background moments; inside safe areas.
- Song audio intact and on loudness target.
- Tell the user: line count, any lines you split, files.
Example requests
- "Make a lyric video for my song with this cover art, karaoke-style highlight."
- "Lyrics on screen over my performance footage, bigger text on the chorus."
- "A 9:16 chorus lyric clip for Reels."
Files (claude-video-editing-skills)
-
SKILL.md 6.6 KB
--- name: lyric-video description: "Lyric-video editor: times every lyric line to the sung vocal (Whisper word timestamps, optionally on a Demucs vocal stem), shows each line with a karaoke fill as it is sung and the next line dimmed, over artwork, footage or a beat-reactive background, and delivers a 16:9 lyric video, a 9:16 chorus clip and an .srt of the lyrics. Use when the user says: lyric video, put the lyrics on screen, karaoke video, sing-along video, animated lyrics, lyrics for my song, visualiser with lyrics. Lyrics must be the artist's approved text. Default skin neon-orbit." license: MIT --- # Lyric Video The words land exactly when they are sung, readable on any background, in the artist's look. - **Best for:** song releases, sing-alongs, devotional and worship lyrics, karaoke versions - **Inputs:** the song (master), the approved lyrics text, cover art / background footage - **Default skin:** `neon-orbit` (use `--no-matte`) · alternatives `paper-cut`, `swiss-poster` (`--no-matte`), `minimal-mono` - **Formats:** 16:9 1920x1080, 9:16 chorus clip (15-30 s), lyrics `.srt` - **Duration target:** the song; genre preset `lyric-video` ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = `video-studio-core`, `SKILLS` = the folder holding both. - First time: `python CORE/scripts/check_env.py`. Preset: `python CORE/scripts/editor.py genre lyric-video`. - Get the lyrics as text, exactly as the artist writes them (spelling, line breaks, censoring). Whisper only supplies *timing*; the words on screen come from the approved text. - Vocal stem (optional, much better timing on dense mixes): `pip install demucs`. ## How a pro cuts this | section | what happens | | --- | --- | | intro | title + artist on the artwork until the first sung word | | verses | one line at a time, low-to-mid frame, calm motion | | chorus | bigger type or a different layout; the same design each chorus so it feels like a hook | | bridge | a visual change (colour, background) | | outro | title card again, credits | - **Timing:** each line appears 0.1-0.2 s before its first sung word and leaves 0.2 s after the last; words fill as sung. No line on screen < 1 s. - **Line length:** <= 42 characters (16:9), <= 26 (9:16); split long lines at the singer's breath. - **Readability:** contrast over any background (shadow/box/dim the background with `vignette`); avoid the face if footage has a singer. - **Motion:** gentle (`maskUp`, `fade`); beat pulses on the chorus only; never shake text people must read. - **Audio:** the song untouched (`"audio": {"voice": false}`), normalised to -14 LUFS. - **Colour:** follow the cover art; a look on footage only (`music-video-neon`, `dreamy`, `vintage-70s`). - **Client (artist) judges:** sync, readability, matches the release artwork. - **Common mistakes:** lyrics typed from Whisper (wrong words), lines late by half a second, two lines changing at once, tiny type, text over the singer's mouth. ## Make the edit 1. A background video carrying the song (artwork still; or footage cut to the song with `timeline.py` / the `music-video` skill): ```bash ffmpeg -loop 1 -i cover.jpg -i song.wav -r 30 -pix_fmt yuv420p -c:v libx264 -tune stillimage -c:a aac -shortest lyr_src.mp4 ``` 2. Plan with word timing from the song: ```bash python CORE/scripts/run.py lyr_src.mp4 --style SKILLS/neon-orbit --genre lyric-video --no-cut --no-matte --format 16:9 \ --model small --prompt "distinct words from the lyrics" --work lyr --stop-after plan ``` 3. If the timing is poor (busy mix), time the vocal stem instead, in its own folder, and use its words for timing: ```bash python CORE/scripts/stems.py song.wav --out lyr_vox/stems ffmpeg -loop 1 -i cover.jpg -i lyr_vox/stems/vocals.wav -r 30 -pix_fmt yuv420p -c:v libx264 -c:a aac -shortest lyr_vox/vox.mp4 python CORE/scripts/prep.py lyr_vox/vox.mp4 --work lyr_vox && python CORE/scripts/transcribe.py --work lyr_vox --model small # read lyr_vox/transcript.json for line start/end times (song time = video time, both start at 0) ``` 4. Write the `lyric` element from the approved text + those times (next section), then: ```bash python CORE/scripts/render.py lyr/plan.json --frames 10,30,60,90 python CORE/scripts/render.py lyr/plan.json python CORE/scripts/finish.py lyr/plan.json --platform youtube ``` Always pass `--format` matching the project: run.py and make_plan.py otherwise use the skin's own format (many skins default to 9:16). ## Refine the plan - **Remove** the skin's `captions` element. - **Lyric element:** `{"type": "lyric", "start": 0, "end": 212.0, "y": 760, "w": 1500, "lines": [{"text": "Seek it anywhere", "start": 31.6, "end": 33.0}, {"text": "you will find me there", "start": 33.2, "end": 35.4}]}` - text from the approved lyrics, times from the transcript (start 0.1 s early). - **Per-word fill:** give a line its own `words` (`[{"w": "Seek", "start": 31.6, "end": 31.9}, ...]`) copied from the transcript but spelled as in the lyrics. - **Title / credits:** `titlecard` with artist and song at 0 s until the first line; credits at the end. - **Chorus emphasis:** a second `lyric` element for chorus ranges with a bigger `style` (`{"fontSize": "96px"}`) and `"pulseOn": "beats"` (add `"beats"` from `beats.py song.wav --out lyr/beats.json` via `make_plan.py --beats`). - **Background:** `vignette` (`"strength": 0.7`) for contrast; `lightleak` (`"intensity": 0.3`) in the bridge. - **Audio:** `"audio": {"voice": false}`; no `music` block (the song is the source audio). - **SRT of lyrics:** put the approved words with their times into `plan.words` so `deliver.py` writes `captions.srt` from them. ## Deliver ```bash python CORE/scripts/deliver.py lyr/plan.json --to youtube --burn no --out-dir lyr/deliver --no-post python CORE/scripts/qc.py lyr/deliver/final_youtube.mp4 --plan lyr/plan.json --platform youtube ``` - 9:16 chorus clip: a new work folder with `--format 9:16` on the chorus range (`prep.py lyr_src.mp4 --work chorus --start 58 --end 84`, then `run.py chorus/source.mp4 ... --format 9:16`), lines <= 26 characters. - Give the artist `captions.srt` (also usable as synced lyrics on platforms that accept SRT). ## Check before delivering - Every word matches the approved lyrics (diff your `lines` against the lyric sheet). - Sync checked at the first line, each chorus start and the last line (+-0.1 s). - Readable on the darkest and brightest background moments; inside safe areas. - Song audio intact and on loudness target. - Tell the user: line count, any lines you split, files. ## Example requests - "Make a lyric video for my song with this cover art, karaoke-style highlight." - "Lyrics on screen over my performance footage, bigger text on the chorus." - "A 9:16 chorus lyric clip for Reels."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.