video-translation
Skill23 starsChanged 41 days agoMITView on GitHub
Translate existing video speech and produce an optional speaker-guided dub with exact source/VAD/transcript joins, local stems, bounded timing, executable technical QA and separate unresolved listening review.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- NoizAI/skillstheir own version, first seen Mar 8, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- NoizAI/skills★ 526same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository30 other files it gives its agents.
- AGENTS.mdAGENTS.md
- src/AGENTS.mdAGENTS.md
- tests/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- av-eventsSkill
- creative-concept-designSkill
- educational-explainerSkill
- ffmpeg-productionSkill
- footage-editSkill
- gemini-visualizeSkill
- gr-advisorSkill
- hardware-evidence-captureSkill
- image-generationSkill
- mlflow-tracesSkill
- movie-commentarySkill
- plugin-maintenanceSkill
- qwen-image-integrationSkill
- qwen-video-integrationSkill
- research-brief-builderSkill
- research-visualization-blenderSkill
- research-visualization-freecadSkill
- reverse-search-video-frameSkill
- spatial-video-analysisSkill
- tts-productionSkill
- video-explainerSkill
- video-generationSkill
- video-productionSkill
- video-researchSkill
- video-to-skillSkill
- weaviate-setupSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Read the integration contract for operator configuration, strict schemas, limits, source mappings and readiness.
- Distinctive: line 41
Use independently configured isolated ChatCut omni_call for speech, visible subtitles, speakers and approximate absolute times.
- Very specific: line 50
Do not invent OCR when subtitles are absent.
- Distinctive: line 66
Never select another speaker or use real enrollment/registry operations.
- Very specific: line 69
Call validate_video_translation_plan.
- Distinctive: line 76
Keep include when uncertain.
- Distinctive: line 92
After a listening issue, pass regenerate_segment_ids for only the affected known groups.
- Distinctive: line 107
Call validate_video_translation_delivery.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
Video translation
Installed workflow resources
For a Claude installer layout, use the adjacent managed support directory
../video-research-resources/ and its ../video-research-resources/docs/integrations/qwen-dubbing.md.
The repository-relative references below apply when using a source checkout or unpacked npm package. Installing resources does not activate optional runtimes, providers or external source components.
Inspect one durable project with get_video_translation_state. Follow the first applicable route: validate/listen to a current delivery, resume a validated plan, translate accepted evidence, or prepare/analyze a new source. Read the integration contract for operator configuration, strict schemas, limits, source mappings and readiness.
The complete workflow is source video → local stems/VAD → reconciled transcript → spoken translation groups/references → local dubbed render → technical QA → separately supplied listening review. Preparation/rendering are executable; source interpretation and translation remain agent work. The external service and isolated Qwen ChatCut must be independently licensed/configured. Missing or unready service is terminal readiness evidence. Never auto-download/launch models, select a credit/provider fallback or treat readiness as submission/voice authority.
Source analysis
With explicit authority for the source and external operations, call check_dubbing_service and prepare_video_translation_project using the actual SHA256, project directory and languages. Preparation extracts full PCM, invokes separate_dubbing_audio and detect_dubbing_speech through the exact contracts, and stores returned audio locally. Never use server-local paths. Pass the requested analysis_only, translation_only, full or resume target and retain the user-approved style_brief. Standalone separation supports the source model selector; detect_dubbing_speech accepts strict VadSettings for the verified threshold/hop/minimum-run/padding controls. synthesize_dubbing_speech executes one approved utterance/reference pair and returns local PCM with unresolved listening.
Use independently configured isolated ChatCut omni_call for speech, visible subtitles, speakers and approximate absolute times. The documented entry point is pinned to Qwen-MM-Plugins 07736672525443c7f8a3f6405eed37d2236f023f; installation and availability remain unknown until checked. Do not silently change providers. At duration <=600 seconds, request the complete source once; longer recordings require ordered bounded windows, complete coverage and reconciled overlaps.
Reconcile audio, visible subtitles, VAD, continuity and audible boundaries. VAD is speech presence, not speaker/sentence authority; low overlap is a review flag. Do not invent OCR when subtitles are absent. Write Transcript to analysis/transcript.json with source/VAD hashes, detected language, evidence_windows, ordered source segments and unresolved_issues. Material unresolved facts block rendering. Analysis-only stops after transcript review.
Translation and per-group references
Author TranslationPlan in plan/translation_plan.json. Copy exact source/VAD/ transcript hashes and languages. Preserve meaning, names, tone and natural spoken delivery. Consume each transcript segment exactly once and in order; merge adjacent same-speaker items when appropriate, recording exact IDs/text/span and a concrete reason. Never merge across speakers or hide omitted meaning with speed.
Choose an evidenced reference independently per group: its own clean speech first, preferably >=1.5 seconds; adjacent same-speaker items with gaps <=1.2 seconds next; nearby clean same-speaker speech if needed. Record IDs, interval and reason. Never select another speaker or use real enrollment/registry operations. The service retains its default emotion settings.
Call validate_video_translation_plan. Duration estimates are only diagnostics. Revise only the affected translation/group/reference when evidence is unresolved or speech cannot fit. Translation-only stops after validation.
Render, resume and listening
Review the complete background stem; omit it only after a whole-stem decision. Keep include when uncertain. Call render_video_translation with the valid plan. It cuts exact references, uses journaled TTS candidates, preserves short speech pace, borrows at most 0.3 seconds/half adjacent gaps, centers silence/fades and accelerates at most 1.18x. Known overlong speech permits at most three candidates for that group when the estimate is <=1.35x; cached candidates count toward the total. Unknown/interrupted effects stop for reconciliation without another request.
Voices retain their source timeline. The full background is mixed when selected with normalize=0; measured two-pass normalization is applied once to the final mix. Supported source video/subtitle streams are copied into Matroska with lossless dubbed audio. QA checks output identity, all streams, duration, full decode, stream preservation, replaced audio, actual decoded mix and source/stem/ segment accounting. Unsupported streams fail explicitly. Each content-addressed render preserves its outputs/failures; full/current.json selects a report. Resume only content-matching completed effects. Never reset or rename unknown intent to make it runnable again. After a listening issue, pass regenerate_segment_ids for only the affected known groups; keep successful groups unchanged. Unknown effects block even across changed plans and generation names. Exhausted overlong speech requires revised translation. Full source/stem refresh uses a new prepared project with fresh evidence joins, preserving the earlier project.
Surface output/report links, segment count, every timing/risk flag and all unresolved listening segments with time range, translated text and reason. Inspect acceleration >1.12x, fill <60%, short references, merged units, silence, retries, order/overlap, meaning, pronunciation, identity drift and mix balance. Listen end to end even with no automatic flags. Write ListeningReview beside the selected report, bound to output/plan hashes, with all five checks per segment: translation, timing, voice_reference, natural_delivery and mix. Retain FAIL and UNRESOLVED cases.
Call validate_video_translation_delivery. Never manufacture listening PASS from technical QA, duration estimates, model confidence or filenames. Report supplied review assertions separately from independent listening/service/model/native qualification and Root's product acceptance.
Protocol/workflow attribution: QwenLM/Qwen-MM-Plugins at the pin above, Apache-2.0 original code/skills. This is an independently authored integration workflow; no original implementation, launcher or model is copied/bundled.
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
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.

