film-dialogue-voiceover
Skill2 starsChanged 51 days agoApache-2.0View on GitHub
Prepare character dialogue and narration before video generation. Recommend ElevenLabs MCP, resolve stable voice IDs for every speaker, generate and verify speech files, measure their timing, and attach them alongside visual references through the configured video provider's supported audio-input interface. Use whenever a film, cutscene, product video, or short includes spoken lines or needs consistent voices and reviewed lip sync.
- Reads credentials
In this repository74 other files it gives its agents.
- card-gameSkill
- pixel-art-spriteSkill
- rhythm-gameSkill
- side-scrollerSkill
- visual-novelSkill
- boss-modelSkill
- character-modelSkill
- environment-sceneSkill
- vfx-effectSkill
- weapon-propSkill
- worldview-3d-asset-production-routeSkill
- film-action-combatSkill
- film-character-sheetSkill
- film-directionSkill
- film-end-creditsSkill
- film-generate-reviewSkill
- film-locationSkill
- film-pipelineSkill
- film-prop-sheetSkill
- film-screen-captureSkill
- film-scriptSkill
- film-shot-promptSkill
- film-storySkill
- film-style-librarySkill
- anime-action-sceneSkill
- cinematic-filmSkill
- game-cutscene-generatorSkill
- product-demoSkill
- social-media-videoSkill
- worldview-gameplay-experience-studySkill
- worldview-gameplay-video-analysisSkill
- boss-battleSkill
- hero-skill-systemSkill
- horror-chaseSkill
- narrative-choiceSkill
- platformer-levelSkill
- puzzle-mechanicSkill
- roguelike-generatorSkill
- worldview-game-asymmetric-information-cooperationSkill
- worldview-game-barricade-delay-and-route-choiceSkill
- worldview-game-character-handoff-and-shared-evidenceSkill
- worldview-game-chase-route-architectureSkill
- worldview-game-death-loop-persistent-cluesSkill
- worldview-game-driving-horror-divided-attentionSkill
- worldview-game-evidence-based-entity-identificationSkill
- worldview-game-horror-experience-rhythmSkill
- worldview-game-horror-mundane-routine-corruptionSkill
- worldview-game-horror-procedural-duty-and-incidentSkill
- worldview-game-horror-returning-place-escalationSkill
- worldview-game-horror-role-and-identity-pressureSkill
- worldview-game-key-item-backtrackingSkill
- worldview-game-limited-save-riskSkill
- worldview-game-lure-hide-escapeSkill
- worldview-game-observation-gated-stalkerSkill
- worldview-game-perception-distortion-and-trustSkill
- worldview-game-relief-resource-with-hidden-costSkill
- worldview-game-restore-power-under-pressureSkill
- worldview-game-roaming-stalker-pressureSkill
- worldview-game-safe-room-pressure-resetSkill
- worldview-game-scarce-inventory-triageSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
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 25
Keep speech, environmental sound, and music separate in the plan and review.
- Distinctive: line 52
If speech is requested but no recording or usable speech service is available, explain the missing setup before a voice-dependent video …
- Distinctive: line 73
Use the user's voice choices when given.
- Very specific: line 79
Keep films/<slug>/voice-cast.json or an equivalent existing manifest.
- Distinctive: line 128
Keep distinct bindings for multiple speakers.
- Very specific: line 151
Use the interface's real reference labels.
- Distinctive: line 158
Verify one representative voiced clip before spending on the remaining batch.
- Distinctive: line 161
If a video MCP cannot accept speech audio, report that before submitting.
- Distinctive: line 171
Check the downloaded video's audio stream and listen to every required spoken line.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Film Dialogue and Voiceover
For prescribed dialogue or narration, prepare the speech before generating the corresponding video. Fix who speaks, which voice they use, what they say, and how long the recording actually lasts. Then supply the audio alongside the character and scene references.
A prompt mentioning a line or filename does not attach audio. Wind, footsteps, music, or an audio stream do not establish that the requested speech is present. Keep speech, environmental sound, and music separate in the plan and review.
This collection recommends audio-first production for controlled voices; it does not claim every Seedance interface is silent without external audio. Some providers offer native speech. Check the selected interface and use that alternative only when it fits the user's accepted voice requirements.
1. Connect speech and video services
Recommend the official ElevenLabs MCP, connected in the user's MCP client. Follow the current hosted setup guide and OAuth flow. Inspect the tools exposed to this session: they must be able to select a voice and return speech audio, not just manage conversational agents. Save returned audio locally before temporary download links expire.
ElevenLabs API is an alternative. Keep its key in the client's secret store or
an ignored environment file, normally as ELEVENLABS_API_KEY; never print or
commit it. Its speech API
uses a real voice_id and compatible model_id. Resolve them from the account's
voice and model listings rather than inventing a default voice.
User-configured third-party MCPs, other TTS APIs, and supplied recordings are also valid. Speech and video generation can use different services. Check each connection's access, voice controls, formats, return type, limits, and the user's authorization for paid calls and uploads. A key is not permission to spend or publish private recordings.
If speech is requested but no recording or usable speech service is available, explain the missing setup before a voice-dependent video batch. Offer ElevenLabs MCP/API or a supplied recording; continue script and visual planning meanwhile. A TTS error does not authorize a silent switch to another voice, text-only speech, or a silent deliverable. A requested silent film needs no TTS setup.
2. Keep one voice assignment per speaker
Extract every speaker from the approved script, including the narrator, off-screen characters, announcements, and recorded messages. Give each a stable character ID and record:
| Field | What it fixes |
| Character ID and role | The person or narrator; on-screen, off-screen, or narration. |
| Provider, connection, voice ID | The actual accessible library voice, retained across scenes and retries. |
| Model, language/accent, settings | A repeatable starting point using controls the model supports. |
| Performance direction | Pace, restraint, breath, pronunciation, and changes warranted by this scene. |
| Preview and status | A sample the user accepted, or a selection within their delegated authority. |
| Rights and availability | Permission to use the voice and continued access during production. |
Use the user's voice choices when given. Otherwise compare available previews, choose for the brief and clarity between speakers, and make a short sample within budget. A display name such as “Adam” is not a universal API ID. Avoid casting from gender or age words alone. Use only authorized clones; setting up TTS does not authorize cloning a real person.
Keep films/<slug>/voice-cast.json or an equivalent existing manifest. This
example is a project record, not an API payload:
{
"version": 1,
"speakers": [
{
"character_id": "mira",
"role": "on-screen",
"provider": "elevenlabs",
"connection": "configured-mcp",
"voice_id": null,
"model_id": null,
"language": "en",
"direction": "Quiet and direct; leave a breath before the final word.",
"settings": {},
"preview_file": null,
"status": "selection-pending"
}
]
}Resolve null fields from actual service results before generation; add each other speaker separately. Reuse the ID for whispers, shouts, and later scenes, changing supported delivery settings instead of recasting. Report an unavailable voice and resolve its replacement before affected lines proceed.
3. Generate and measure the lines
Give each utterance a line ID and exact text. Generate one file per line or an intentional same-speaker passage. Preserve approved words, pronunciation, pauses, and delivery. Keep these speech files free of music and background effects; retain a master when converting to a provider-compatible format.
Download and decode the output. Listen for omitted or added words, wrong names, effects, silence, clipping, and changed voices. Measure duration with available media tools: a word-count estimate is not a measurement. Record the script revision and voice/settings so a changed line invalidates only affected audio and shots. Reuse accepted recordings across alternate picture takes.
In speech-lines.json or an existing manifest, record speaker ID, exact text,
file, measured duration, target block/shot, start offset, on-screen mouth or
off-screen role, provider input slot, and review status. Fit the shot around
the recording plus entry, response, and breathing room. Split or revise long
passages within scope instead of cutting final words or silently speeding up
the performance.
Keep distinct bindings for multiple speakers. If the video interface accepts only one audio track, build a timed dialogue track from the individual files and retain a speaker-turn map, or split the exchange into shots. Preserve pauses; overlapping speech needs explicit timing and a provider that can handle it.
4. Attach speech to the video request
Inspect the selected website, MCP schema, or API before the first voiced take. Record whether audio can guide the performance/timing, supply voice style only, or cannot be input at all. A timbre reference alone does not guarantee the given words or waveform will appear in the output.
For Seedance 2.5, use its interface's supported audio-reference workflow; consult the current official input documentation when using the official API. Third-party wrappers can expose different fields, limits, upload methods, and output-audio controls for the same model. Model access alone does not establish audio-input support.
- Attach the actual speech file through a supported upload, asset ID, or provider-fetchable URL, alongside required image/video references. Check format, size, duration, and reference limits. Prefer authorized asset storage or scoped links; do not publish a private recording just to satisfy a wrapper.
- Bind each audio input to its speaker, character reference, line, and shot.
Use the interface's real reference labels. A guessed
{line} [Audio1]prompt oraudio_referencesfield is not a portable replacement for an attachment. - Enable the supported speech/audio output mode. Name which mouth follows which recording, who stays silent, and which lines are narration. Narration should not animate a visible character's mouth.
- Inspect submitted arguments or website attachments: audio must survive the wrapper, not just the images. Save the task ID and a secret-free input record.
- Verify one representative voiced clip before spending on the remaining batch. Check words and speaker assignments as well as the picture.
If a video MCP cannot accept speech audio, report that before submitting. Use
another already-authorized compatible connection, or agree on an alternative.
Narration can be mixed over footage later using the prepared recording; visible
speech may need a supported lip-sync pass or a new shot. Do not present that as
audio-conditioned generation. The bundled image/parameter-only
film-pipeline/scripts/batch.sh is not an audio uploader; use a verified
audio-capable MCP/API route for these takes.
5. Review the speech, then mix
Check the downloaded video's audio stream and listen to every required spoken line. Verify words, voice identity, timing, intact endings, no extra speaker, and audible level. Review visible mouths during playback; metadata or a transcript cannot establish lip sync. If listening or playback is unavailable, report that gap rather than claiming the take passed.
Music, ambience, effects, and final leveling can follow picture lock. Preserve accepted speech and timing. Avoid doubling speech by adding the original TTS track when it is already present. When replacing speech, remove or mute the old voice and preserve other sound with available stems or a deliberate remix. Narration repair can often retain the picture; replacing audio alone does not repair the timing of a visibly speaking face.
A missing line, switched voice, or failed attachment fails the voiced take even if the task completed. Correct that cause and retry only within budget. Deliver the video, cast and line manifests, retained speech files, and review notes. Report measured results, never “perfect lip sync” or unsupported percentages.
Worked handoff
In an original scene, Mira says “Keep the lower gate closed,” Ivo replies “Then take the stairs,” and an off-screen narrator closes the scene. Select three voices and retain their real library IDs.
Suppose the files measure 2.6, 2.1, and 3.4 seconds; these are example values, not promised timings. Place Mira at 0.5–3.1 s, leave a 0.8 s pause, place Ivo at 3.9–6.0 s, and the narrator at 6.8–10.2 s over the departure. Choose a supported clip length with room for the ending beat, or split shots.
Attach the character references and speech inputs, or the timed dialogue track with its turn map. Each character mouths only their own line; neither mouths the narration. Check this clip before using the cast for more scenes. This is a handoff example, not a claim that TTS or video generation has been run.
Related Skills
- Film Shot Prompt: uses recordings and measured timing before constructing a voiced request.
- Generate and Review: submits verified inputs, checks the speech, and assembles accepted takes.
- Cinematic Film and Game Cutscene: route spoken scenes here before spending the video-generation budget.
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.

