For ordered independent edits of an existing clip, use scenario-ad-variants; this skill starts new creative.
This skill is the director layer: it turns one product shot and a brief into a finished ad, delegating mechanics to the sibling skills it names per stage. Connection and the core loop: see the scenario skill in this repo. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap. The full sibling map and install commands: references/dependencies.md.
Two rules bind every stage, because they carry the ad's credibility. The product is never generated: every product shot starts from the uploaded photo, as a first frame or composited into an approved still. Text is never generated: logos, taglines, prices, and legal lines are overlaid in assembly on text-free plates.
\nAsk the brief once, then run without stopping. The one approval that matters sits between the priced storyboard and generation: video is the expensive stage, so the user signs off on the budget table first. The concept page and the stills are cheaper checkpoints: show them when someone can answer. In a run where nobody can, continue: the brief's spend ceiling stands in for the sign-off, so generate video only while the priced board fits it, and the fidelity gate stands in for still approval.
\n| Stage | \nWhat happens | \nDetail | \n
|---|---|---|
| 1. Brief | \nOne objective, audience, proposition, creativity and length dials, platforms, mandatories, spend ceiling | \nreferences/brief.md | \n
| 2. Concept | \nOne page, anchored to a named famous-ad pattern, shown to the user | \nreferences/storyboard.md | \n
| 3. Storyboard | \nBeats and panels per the objective's rulebook, priced shot by shot with dry_run, approved | \nstoryboard.md, brief.md | \n
| 4. Stills | \nComposed natively per delivery ratio, product composited from the real photo, approved before animating | \nscenario-image, scenario-consistency | \n
| 5. Animate | \nImage-to-video per approved still, one camera move, generateAudio: false where the schema has it, wait=false | \nscenario-video, scenario-seedance | \n
| 6. Fidelity gate | \nExtract frames from every product or character clip, spot-the-difference against the approved reference | \nstoryboard.md, scenario-asset-analysis | \n
| 7. Audio | \nMusic generated on the cut's beat grid, voiceover to a word budget, sonic logo | \nscenario-audio, references/delivery.md | \n
| 8. Assemble | \nTimeline, text cards from scenario-text-overlay as image layers, captions in safe zones | \nscenario-video-assembly, delivery.md | \n
| 9. Pre-flight | \nPolicy, flashes, sound-off pass, loudness, spend report | \ndelivery.md | \n
asset_analyze, instructing it to inventory label text, cap geometry, and glass tint: every later check reuses it (asset_describe returns a style line, not this inventory).dry_run every planned run, multiply by retry floors, present the budget table.model_schema_get first, prompt movement first (\"slow push-in over 4 seconds, the bottle rests fixed in place\"), generateAudio: false when the schema lists it.asset_analyze call against the checklist. Label drift fails the clip; regenerate from the same still, never from the drifted output.scenario-text-overlay renders the tagline card into the safe zone; caption the master.usage against the approved board.model_run on Scenario tool models (see scenario-video-assembly).generateAudio, where offered) at its default: the soundtrack is built in assembly, so clips with baked-in audio fight the bed.