feat(seedance): production QA section — audio-vs-script drift (post-gen verification)#21
Merged
Merged
Conversation
…on (post-gen verification) Backfills a real dogfood insight from local module use into the OSS library. When `generate_audio: true` is enabled (the default on Seedance 2.0) and your content has dialogue, Seedance produces real spoken audio but the actual words spoken may diverge from the dialogue you wrote in the prompt. The model is creative, not literal — prompt dialogue in double quotes is a strong hint but for longer / narrative content the model can omit lines, rephrase, substitute homophones, or fill silences with ambient sound. This matters for subtitled-content publish workflows; doesn't matter for vibe / no-dialogue clips. Body adds (between "Trusted-result chaining" and "Polling vs callback") - 6-step production pattern: generate → ffmpeg extract → TOS upload → Seed-ASR transcribe → diff vs target → tag mismatches for editorial - 5-row status taxonomy mirrored from `library/volcengine-speech`: asr_ok_text_match / asr_ok_text_mismatch / video_incomplete / asr_unreliable / asr_no_speech (consistent label convention across the trove content pipeline) - 3 production rules: don't silently overwrite target subtitles with ASR; subtitles are a post-production layer; `status: "succeeded"` is not content QA — audio-vs-script is a separate gate - Cross-references to the volcengine-tos (audio hosting) and volcengine-speech (transcription) modules — closes the production loop with explicit links Privacy - Maintainer's local module had this insight in a "Production Note (2026-05-22)" section with private references (downstream project name, episode ID, file name, internal field/tool names). All stripped — the OSS version frames the same engineering insight in general terms with public placeholders - Pre-commit hook PRIVATE_RE scan: clean on staged diff Validation - bun bin/cli.ts validate library/seedance: 0 errors / 0 warnings Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Backfills real dogfood from local seedance module: when `generate_audio: true` (Seedance 2.0 default) and content has dialogue, the model produces real spoken audio but the actual words spoken may diverge from the dialogue in the prompt. For subtitled publish workflows this is a silent production trap.
What lands
New section between "Trusted-result chaining" and "Polling vs callback":
Privacy
Local original had this in a "Production Note (2026-05-22)" section with private references (downstream project name / episode ID / file name / internal field+tool names). All stripped; OSS framing uses public placeholders and engineering shape only.
Verification
🤖 Generated with Claude Code