agentleFS
godot-fungodot-fun/gai/.agents/skills/storyboard-tts/SKILL.md

storyboard-tts

Skill210 starsChanged 2 days agoMITView on GitHub

Converts storyboard markdown (bilingual Chinese + English narration per shot) into speech audio via IndexTTS2 (same stack as ai-text-to-speech). Batch-writes WAVs under Chinese/ and English/ named by shot id (model loaded once), pads 0.4 s edge silence in place, and can generate speech-timeline.md plus one concatenated SRT per language (Chinese.srt / English.srt). Use when the user wants storyboard TTS, storyboard to speech, narration VO, storyboard-to-speech, bilingual VO export, subtitles, or batch TTS from a storyboard.md.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 1 is distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    When this skill applies, read and follow dependency-manager — run scripts as documented, install missing tools into .dependency/.

    line 31
  2. Very specific:

    Never overwrite the storyboard source.

    line 36
  3. Very specific:

    Write Chinese/<id>.wav and English/<id>.wav (skip existing unless --force)

    line 97
  4. Very specific:

    Write speech-timeline.md and Chinese.srt / English.srt when --report

    line 99
  5. Distinctive:

    Do not invent narration — use storyboard Chinese/English fields as-is (audio and subtitles).

    line 156
  6. Very specific:

    Prefer one synthesize.py invocation for a full board.

    line 157
  7. Very specific:

    If audio already exists and only subtitles are needed, run write_subtitles.py alone (stdlib python).

    line 162

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Storyboard TTS

Take a storyboard deliverable and batch-synthesize Chinese + English voice-over with IndexTTS2 (shared setup with ai-text-to-speech).

OutputPath
Chinese VO<storyboard-dir>/<voice-stem>/Chinese/<shot-id>.wav
English VO<storyboard-dir>/<voice-stem>/English/<shot-id>.wav
Duration doc<storyboard-dir>/<voice-stem>/speech-timeline.md
Chinese subs<storyboard-dir>/<voice-stem>/Chinese.srt
English subs<storyboard-dir>/<voice-stem>/English.srt

Shot id from headers (### Shot 01 — … → 01.wav).

Subtitles: one SRT per language. Shots are laid end-to-end on the VO timeline (shot N starts when N−1 ends). Inside a shot, text is split on sentence punctuation (.!?;… and CJK equivalents) into multiple cues; cue lengths share that shot’s WAV duration by non-whitespace character weight. Skip (no VO) / missing audio.

Rules

When this skill applies, read and follow dependency-manager — run scripts as documented, install missing tools into .dependency/.

  1. Batch synthesis only via .ai/storyboard-tts/synthesize.py with the index-tts interpreter. Do not hand-write IndexTTS loops, temporary batch drivers, or N× single tts.py calls for a full storyboard.
  2. Trial / single-line checks may use ai-text-to-speech tts.py, or synthesize.py --limit 1.
  3. Parse-only / report-only / subtitle-only steps use stdlib python (.dependency/python/python).
  4. Never overwrite the storyboard source. Write only under <audio-dir>/.
  5. Skip (no VO) / empty lines — no empty WAVs or empty subtitle cues.
  6. Confirm voice reference before a full batch. Confirm the output directory only when the default would be ambiguous or collide with unrelated output.
  7. The top-level output directory uses the shared reference voice filename stem. With separate references and no --voice, it uses the Chinese reference stem for both/chinese, or the English stem for english. Subdirectory names stay fixed. Default <audio-dir> is <storyboard-dir>/<voice-stem>/. Do not use <storyboard-stem>-speech.
  8. For a normal full delivery, include --report so the timeline and SRT files promised by this skill are written. Omit it only when the user wants WAVs alone.

Inputs

RequiredNotes
Storyboard .md### Shot NN — title with - **Chinese:** / - **English:**
Reference voiceWAV/MP3 for IndexTTS (--voice, or --voice-zh / --voice-en)
OptionalDefault
Output dir<storyboard-dir>/<voice-stem>/ (reference audio filename, no extension)
Languageboth (--lang chinese / english)
--fp16 / emotion / --devicesame meaning as ai-text-to-speech
--forceoff (skip existing WAVs)
--limit N0 = all jobs (use 1 for trial)
--reportoff by default in the script; enable for the normal full delivery to write speech-timeline.md + Chinese.srt / English.srt
--no-subtitleswith --report, skip SRT files
Edge padon (0.4 s); --no-pad / --pad-duration / --pad-threshold

Layout

<storyboard-dir>/<voice-stem>/   # e.g. narrator-self-fast/ from narrator-self-fast.wav
  Chinese/
    01.wav
    …
  English/
    01.wav
    …
  shots.json
  speech-timeline.md
  Chinese.srt            # all Chinese cues, continuous timeline
  English.srt            # all English cues, continuous timeline
  _text/                 # only with --write-text

Quick Start

From project root:

.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --storyboard path/to/storyboard.md --voice path/to/ref.wav --fp16 --report

Trial run (first line only):

.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --storyboard path/to/storyboard.md --voice path/to/ref.wav --fp16 --limit 1

Writes under <storyboard-dir>/<voice-stem>/ (override with --audio-dir only when needed).

This will:

  1. Parse the storyboard → <audio-dir>/shots.json
  2. Load IndexTTS2 once
  3. Write Chinese/<id>.wav and English/<id>.wav (skip existing unless --force)
  4. Pad each WAV in place to 0.4 s leading/trailing silence (--no-pad to skip)
  5. Write speech-timeline.md and Chinese.srt / English.srt when --report

Separate voices / language

.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --storyboard path/to/storyboard.md --voice-zh path/to/zh_ref.wav --voice-en path/to/en_ref.wav --fp16 --report
.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --storyboard path/to/storyboard.md --voice path/to/ref.wav --lang chinese --fp16 --report

Parse, report, or subtitles alone (stdlib python)

.dependency/python/python .ai/storyboard-tts/parse_storyboard.py path/to/storyboard.md -o path/to/<audio-dir>/shots.json
.dependency/python/python .ai/storyboard-tts/duration_report.py --storyboard path/to/storyboard.md --audio-dir path/to/<audio-dir> --shots path/to/<audio-dir>/shots.json -o path/to/<audio-dir>/speech-timeline.md
.dependency/python/python .ai/storyboard-tts/write_subtitles.py --audio-dir path/to/<audio-dir> --shots path/to/<audio-dir>/shots.json

Resume from an existing shots.json:

.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --shots path/to/<audio-dir>/shots.json --voice path/to/ref.wav --fp16 --report

Common Flags (synthesize.py)

FlagNotes
--storyboard / --shotsSource (one required)
--audio-dirOutput root (default: <storyboard-dir>/<voice-stem>/)
--voiceShared speaker ref
--voice-zh / --voice-enPer-language refs
--langboth (default), chinese, english
--limit NFirst N pending jobs only
--forceRe-synthesize and overwrite existing WAVs
--reportWrite timeline + SRT subtitles
--report-outCustom timeline path
--no-subtitlesSkip SRT when using --report
--write-textDump lines under _text/
--no-padSkip in-place 0.4 s edge padding (off by default — padding is on)
--pad-durationTarget silence per edge in seconds (default: 0.4)
--pad-thresholdSilence detect threshold in dB (default: -50)
--fp16 / --deviceRuntime
--emotion-* / --random / --verboseSame role as tts.py

On partial failure: script continues remaining jobs, prints Failed jobs: …, exit code 1. Fix install/voice per ai-text-to-speech troubleshooting, re-run (existing OK files are skipped).

Agent Notes

  1. Do not invent narration — use storyboard Chinese/English fields as-is (audio and subtitles).
  2. Prefer one synthesize.py invocation for a full board; model reload cost is the reason.
  3. Chat summary: audio-dir (<storyboard-dir>/<voice-stem>/), counts, path to speech-timeline.md, Chinese.srt / English.srt, Chinese/English total seconds — no full transcripts unless asked. Omit --audio-dir unless overriding.
  4. IndexTTS install lives in ai-text-to-speech; do not duplicate Setup here beyond “populate index-tts if missing”.
  5. Edge padding is built in (default 0.4 s, in place via pad.py). Use --no-pad only when they want raw TTS with no extra silence; --pad-duration if they want a different length.
  6. Loudnorm / OGG / trim remain separate skills after this one.
  7. If audio already exists and only subtitles are needed, run write_subtitles.py alone (stdlib python). Re-running synthesize.py --report without --force skips synthesis but may still modify existing WAVs to satisfy edge padding, then rewrites the timeline/SRT. Use --no-pad when existing WAV bytes must remain untouched.

Tests

Stdlib scripts (from repo root):

.dependency/python/python .ai/storyboard-tts/test_parse_storyboard.py
.dependency/python/python .ai/storyboard-tts/test_write_subtitles.py

IndexTTS batch driver (requires populated index-tts; see cli/storyboard-tts.md):

.dependency/index-tts/.venv/Scripts/python.exe .ai/storyboard-tts/synthesize.py --storyboard path/to/storyboard.md --voice .ai/test/audio/han.wav --fp16 --limit 1 --report

Related

  • storyboard — source markdown
  • ai-text-to-speech — single-line TTS + IndexTTS setup
  • storyboard-av-mix — mux VO with per-shot video
  • Optional after: audio-loudness-normalization

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.