scenario-kinetic-film writes a film as spec.js: scene blocks (Words, Strike, Graph, Prompt, Picture, Logo and others) placed on the music's beat grid, plus a palette, captions and cue list. Its engine renders that as one page that is a pure function of time, and its Python scripts render, mix and gate the result (film-as-function-of-time). Scenario MCP makes what code cannot draw: a music bed at an exact tempo, a kit of about 12 one-shot effects, and the key art the Picture block cuts in. Route a beat-cut, type-led teaser or hype reel to it when it is installed or the user names it. A HyperFrames film stays with HyperFrames and a Remotion film with cinetic: one engine per project.

Who owns what

scenario-kinetic-film owns	This graph owns
`spec.js`, the blocks, the engine, the render (`render.py`), the mix (`mix.py`) and `finish.py`'s gates	the register (launch-video-registers) and the beats (launch-video-structure)
the Scenario loop: `recommend` → `model_schema_get` → a `dry_run` price → the user's go → run → `jobs_wait` → `asset_download`	type cadence (launch-video-type), seams and cuts (launch-video-seams, launch-video-cuts)
prompting key art and the sound kit	where sounds go, subtraction and loudness (launch-video-sound, sound-motion-sync)

Its register is not this graph's kinetic-type row

Its references are high-energy motion-design reels: the fastest is 16 % still, has a longest still of 0.37 s, and cuts 9.1 times per 10 s (55 a minute). Kaiju Postal, its own teaser, cuts 75 a minute. That sits at or past the fast edge of this graph's beat-cut sizzle row (30.7 a minute, up to 60.6), not in its kinetic type row, whose three films median 3.3 cuts a minute. So finish.py's gates (≤ 40 % still, no still over 0.6 s before the end card, median motion ≥ 2.5 px a frame) are right for that register and wrong for any other. A product-camera film is 49 % still and would fail them.

Its still share is not this graph's still share either: it counts a frame as still when under 1.5 % of a 320 × 180 frame changes, while launch-video-review counts one only when nothing visible changes. Compare its numbers with its own references only. The references are unnamed and uncounted, so quote them as "scenario-kinetic-film reports", the same qualifier cinetic-reconciliation uses.

Where this graph's values win

Topic	scenario-kinetic-film	This graph
Fades	"the tell of template motion"; only the logo card's small lines	11 of 29 Skale and 6 of 18 acclaimed films dip or dissolve; only a dip through a ground neither scene has fails (launch-video-cuts)
Sound on cuts	"every cut has a sound"	subtraction punctuates: a pulled sub or a silence before a reveal in 23 of 28 Skale films (launch-video-sound)
Loudness	`mix.py` masters to −14 LUFS; the gate accepts −16 to −12	a register choice; medians −18.1 and −18.4. −13 suits this register, because three of the four beat-cut sizzles sit at −12 to −13
9:16 version	the 16:9 film centred on black	re-lay it out; never crop or letterbox (launch-video-formats)
Type size	300–600 px words on a 1920 frame	right for this register only; statements in the measured films are ~5 % of frame height

What it adds

Outline, then fill: an outlined word for something not yet true fills on its beat (NO TIMELINE.).
Runs of beat cuts accelerate (1, 1, 1, ½, ½, ¼ beat, then a flash), and a run can flip the ground on every cut.
A custom sound for the film's own joke (a knock, a footstep, a bell) earns more than another stock impact (designed-sfx-anatomy).
Generated key art: describe the composition and never ask for empty space; darken for type in the film. Inspect every picture at 100 % for real makers' marks before it goes in.

Gotcha

Scenario runs cost money, and a failed or blocked run can still be charged. Price every step with dry_run, give the user the total for the whole soundtrack and picture set, and wait for a go. Take model ids from recommend, never from memory. If teams_list returns several teams or projects, ask which one to use. Never write "no watermark" when a reference image is attached: the request is blocked as watermark removal, and charged. "No text, no logos" is enough.

Sources

Scenario, scenario-kinetic-film skill (zip received 2026-10-30, no licence file: cited, never copied): SKILL.md, references/method.md §1, §5 (energy targets), references/craft.md, references/sound.md, references/scenario-mcp.md, references/blocks.md, examples/kaiju-postal/. Its companion skills: npx skills add scenario-labs/skills; its MCP server is https://mcp.scenario.com.
This graph's values: docs/research/launch-films/ (HKTITAN, 2026-09-28).
Related: skill-router, hyperframes-reconciliation, earshot-reconciliation.