Ular KimsanovandClaude Opus 4.7
a78ec5ec76
* fix(skill): all file paths use capture/ subfolder prefix Step-1 was updated to capture into <project-dir>/capture/ but all file-read instructions across steps 1-6 still used bare paths (extracted/tokens.json, assets/hero.png). Agents following the skill looked for files in wrong locations, producing broken compositions with 404'd images and fonts. - step-1: all 12+ file-read paths prefixed with capture/ - step-2: tokens.json and asset-descriptions.md paths fixed - step-3: visible-text.txt path fixed - step-4: asset-descriptions.md, asset examples, directory tree expanded to show capture/ children (screenshots/, assets/, extracted/) - step-5: minor cleanup - step-6: asset-descriptions.md path fixed, composition examples already correct (../capture/assets/) - techniques.md: Lottie, video, font paths all prefixed Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(skill): address Copilot review — font comment path, add videos/ to tree - techniques.md: comment says ../capture/assets/fonts/ (matches composition context) - step-4 file tree: add videos/ under capture/assets/ --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2.4 KiB
Step 5: Generate VO + Map Timing
Audition voices
Never use the first voice you find. Audition 2-3 voices with the first sentence of SCRIPT.md:
- Kokoro (try first — free, no API key) —
npx hyperframes tts SCRIPT.md --voice af_nova --output narration.wav. Runs locally on CPU. Requires Python 3.10+ (macOS system Python 3.9 won't work — if it fails with an onnxruntime error, move to the next option). - ElevenLabs (best voice quality, widest selection) —
mcp__elevenlabs__search_voicesto browse,mcp__elevenlabs__text_to_speechto generate. Does not return timestamps — transcribe separately after. - HeyGen TTS (returns word timestamps automatically — saves a transcribe step) —
mcp__claude_ai_HeyGen__text_to_speech. Use when you want timestamps without a separate transcription pass.
Pick the voice that sounds most natural and conversational. Listen for pacing — does it breathe between sentences? Does it sound like a person or a robot?
Generate full narration
Generate the full script as narration.wav (or .mp3) in the project directory.
Also save the exact spoken text — with pronunciation substitutions applied (e.g., API → A P I, $2T → two trillion) — as narration.txt in the same directory. This is the string passed to TTS, distinct from SCRIPT.md which is the human-readable creative doc. Having narration.txt makes it trivial to regenerate the audio later with a different voice without re-deriving the substitutions. Name it exactly narration.txt.
Transcribe for word-level timestamps
npx hyperframes transcribe narration.wav
Produces transcript.json with [{ text, start, end }] for every word. These timestamps are the source of truth for all beat durations.
Map timestamps to beats
Go through STORYBOARD.md beat by beat. For each beat:
- Find the first word of that beat's VO cue in
transcript.json - Find the last word of that beat's VO cue
- Set
beat.start = firstWord.start,beat.end = lastWord.end - Add 0.3-0.5s padding at the end for visual breathing room
Update STORYBOARD.md with real durations. Replace estimated times (e.g., "0:00-0:05") with actual timestamps (e.g., "0.00-3.21s").
Beat boundaries land on word onsets — hard cuts to the VO.
Update index.html
Update each scene slot's data-start and data-duration to match the real beat timings from the transcript. Also update the total composition duration and audio element duration.