Skip to content

gooseworks-ai/goose-skills / skills/video/video-production-harness

video-production-harness

Verified

Full video production sequence with script review, actual ingredient choices, controlled generation, editing, evidence-based quality review, polish, captions and delivery. A host binding supplies project storage, authentic approvals, provider access and billing.

FileKindSize
SKILL.mdskill md6.0 KB
auto-fix-from-review-notes.mdmarkdown6.5 KB
auto-refine.mdmarkdown5.3 KB
brainstorm.mdmarkdown3.2 KB
capabilities.mdmarkdown13.6 KB
create-clips.mdmarkdown15.0 KB
create-design-brief.mdmarkdown13.4 KB
create-storyboard.mdmarkdown8.8 KB
edit-clip.mdmarkdown6.4 KB
edit-video.mdmarkdown12.8 KB
lock-character.mdmarkdown12.2 KB
lock-script.mdmarkdown8.7 KB
orchestrator.mdmarkdown10.9 KB
polish.mdmarkdown9.2 KB
preflight-audit.mdmarkdown6.2 KB
promote.mdmarkdown3.2 KB
references/check-severity.mdmarkdown9.7 KB
references/creative-contract-and-review.mdmarkdown11.4 KB
references/creator-grammar.mdmarkdown6.7 KB
references/editorial-review.mdmarkdown6.4 KB
references/genres/ai-creator-ugc.mdmarkdown2.7 KB
references/genres/chat-ui.mdmarkdown2.2 KB
references/genres/hypermotion.mdmarkdown1.4 KB
references/genres/motion-graphics.mdmarkdown1.5 KB
references/genres/narrator-montage.mdmarkdown3.1 KB
references/genres/product-demo.mdmarkdown2.0 KB
references/hook-compatibility.mdmarkdown2.8 KB
references/libraries-look-and-rubric.mdmarkdown6.0 KB
references/product-scale-and-world.mdmarkdown4.2 KB
references/quality-tier.mdmarkdown9.4 KB
references/recipe-format.mdmarkdown7.1 KB
references/review-script-shortform.mdmarkdown11.7 KB
references/shot-routing.mdmarkdown14.9 KB
references/specialist-handoff.mdmarkdown4.7 KB
references/speech-and-route-qualification.mdmarkdown10.1 KB
references/talking-actor.mdmarkdown6.6 KB
release.jsonconfig22.1 KB
review-video.mdmarkdown15.0 KB
scripts/assemble.pyscript11.1 KB
scripts/assemble_sources.pyscript5.7 KB
scripts/assembly_timeline.pyscript11.4 KB
scripts/check_recipe.pyscript9.6 KB
scripts/final_grade.pyscript2.9 KB
scripts/final_measure.pyscript14.1 KB
scripts/final_measure_parts.pyscript14.6 KB
scripts/pack_recipe.pyscript11.3 KB
scripts/qc_evidence.pyscript15.0 KB
scripts/reference_rate.pyscript4.3 KB
scripts/reference_shots.pyscript13.1 KB
scripts/timeline_rules.pyscript10.8 KB
skill.meta.jsonconfig820 B
states/edit-to-delivery.mdmarkdown8.2 KB
states/setup-and-script.mdmarkdown10.8 KB
states/storyboard-and-clips.mdmarkdown8.1 KB
wrap-session.mdmarkdown5.9 KB

references/talking-actor.md

View on GitHub
# Human version

For realistic people speaking to camera, choose the person, hear the voice reading the script, then check the generated performance. Use three new actor options when casting is needed. Keep the approved format: a phone testimonial, a studio presenter and a fictional character need different looks.

---

# Agent version

## When to use this method

Read this for a generated person speaking on camera, including speaking scenes inside demos or mixed-format ads. It transfers the tested talking-actor specialist method into the existing production phases; it is not another project, endpoint or approval flow. Narration, songs, cartoons, recorded actors and motion-only scenes keep their own routes from [speech qualification](speech-and-route-qualification.md).

Discover the specialist when the host exposes it, and inspect its actual package and dependencies. The standalone version's direct-provider scripts, old prices, image models and local helper paths are not the host contract. Use only supported, project-billed operations, with current quotes and approvals. Do not require a specialist that is absent from the current catalog; the method below fits the existing harness.

## Cast once, then audition

- Reuse a supplied or accepted actor. Otherwise make **three distinct actor candidates**, one still per candidate, within the approved preview budget, and show one contact sheet at the intended viewing size. These are alternative people, not three angle sets. Generate angles only for the chosen anchor and only when a scene needs them.
- Preserve the brief's casting and performance requirements. Match framing to the action: a propped camera can leave both hands available; a selfie cannot. Lock the chosen identity and reuse it across scene stills and takes.
- After the actor choice, shortlist two or three compatible voices from the host's approved catalog. Play the **same approved script excerpt** in each, including a difficult name and a conversational turn. For a short ad, use the whole script when budget permits. Keep model, supported settings and playback loudness comparable. Show playable samples beside the actor, not voice names alone.
- Keep the customer choice in the existing ingredient review. An agent may recommend a voice but does not silently select an unheard one. Reuse a previously approved voice/sample when the words and delivery still fit. Explicit delegation of creative choices can select a voice, but the chosen playable sample still appears in the review; money approval remains the host's.
- Save selected actor/voice IDs, script revision, sample files, settings and provenance. A transcript confirms neither naturalness nor correct pronunciation: hear the actual words when listening is available and record any missing capability.

## Choose the speaking route

For a realistic generated performance needing the selected voice's **character**, prefer joint voice-and-picture generation when the exact endpoint supports audio references, the shot type is qualified and the host can quote and approve that route. This can work line by line in a longer ad; it does not mean putting the whole ad into one take.

An audio reference conditions timbre and delivery; it does not preserve the supplied waveform or exact timing. Record `native_dialogue`, keep the generated take's own sound, and compare it with the audition. If the final must use an exact recording, choose qualified audio-driven lip-sync instead. Narration over a non-speaking face stays narration. Do not add a lip-sync pass or substitute the audition track over a native speaking take.

The tested agent-built H3 Max route uses the scene still and locked character as `reference_image_urls` and the approved voice read as `reference_audio_urls`, with prompt expansion disabled where supported. Inspect the actual endpoint/schema and free quote before sending these fields. A kit part without audio-reference inputs cannot implement this route; use a supported managed operation or present the supported alternative. A Fabric tier quote is not an H3 quote: follow [quality tier](quality-tier.md) and obtain the route's real approval before spending.

For that qualified H3 route:

1. Put scene, action and performance direction before the spoken copy. Keep copy in the endpoint's documented dialogue block, closed with an instruction to stop speaking. Inspect the helper's syntax; do not send provider tags to a wrapper that refuses them.
2. Size each take to its actual line and the endpoint's limits. A short line may need explicit immediate speech onset. Meet the documented audio-reference minimum (the tested route required two seconds); a longer read of the same approved voice is a timbre reference, never a different script to speak.
3. Quote a bounded redo allowance, up to three takes per line, rather than generating three automatically. Inspect the first take before the rest. Keep the least expensive accepted result; reconcile a failed or interrupted job before resubmission.
4. Compare actual words with locked copy, inspect any added tail or internal cut, and hear pronunciation, joins and delivery. A pause over one second is a flag to inspect against the planned performance, not a universal ban on deliberate pauses. Gaze follows the approved direction; a profile conversation need not stare at the lens.
5. Reject unintended words in the retained performance. A discarded tail may be cut only outside the wanted speech, with picture/audio linked; then recheck the edited clip and its checksum. Never trim away a required action or word to pass a check.

## Finish and review in the approved style

Phone finishing belongs to a phone-footage brief. Compare with several appropriate real clips before adding softness, grain, room tone or camera motion, and tune only what the take needs. A steady propped phone need not shake. Studio, cinematic, stylized and animated work keeps its approved light, focus, texture and sound. No universal filter values or invented room sound.

The original comparison found limited head movement across its tested configurations. Treat this as a known risk to inspect, not a guarantee about all future models or a reason to force exaggerated acting. Never promise parity with filmed actors from a score or transcript.

Use the existing [clip checks](../create-clips.md) and [finished-video review](../review-video.md): actor continuity, full speech, mouth motion, pronunciation, performance and the final mix. Compare the generated voice with the approved audition, especially names and voice changes between takes. Label measured evidence honestly; ASR, a clean waveform and file checks cannot establish heard naturalness or perceived sync. Review the exact edited bytes after finishing; preserve accepted assets on repair.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 323kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 323kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 323kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 323kUpdated Claude CodeCodexVerifiedInstall