/cardstock-diorama-myth

name: cardstock-diorama-myth-animation
description: >-
  Produce a complete narrated MYTHIC SHORT FILM end-to-end on Unsora MCP in the
  layered cut-paper diorama style — stacked matte cardstock worlds, a strict three-colour
  palette, exposure that swings, paper wipes hiding every seam, landing on a face.
  Animation-movie storytelling, not an explainer: gods, monsters, heroes, folklore, origin
  myths and transformations, told in ten-second story beats with a mythic narrator, native
  paper foley, and one assembled MP4. Use whenever the user wants a myth, legend, fable,
  folktale or fairy-tale short, a cut-paper or papercraft animated story, "tell the story
  of X" as a video, a stop-motion-look mythic film, or says "run the cardstock diorama
  myth animation" or "run the diorama myth pipeline" — even with no story given, since the skill proposes myths itself. NOT for
  fact-driven explainers with charts and statistics (that is vox-motion-graphics).

Cardstock Diorama Myth Animation (Unsora MCP)

One request — a myth, a premise, or nothing at all — becomes a finished narrated mythic short film in the layered cut-paper diorama style: stacked matte cardstock worlds, three colours and no others, a storyteller's voice, one final MP4. This is **animation-movie storytelling** — gods, monsters, thresholds, transformations — not a data explainer.

**The style, in one line:** flat matte cardstock worlds — visible cut edges, soft contact shadows, a strict **three-colour palette and no others** — a tiny figure dead-centre of a huge symmetrical set, **rendered in 3D, not real stop-motion**, scenes packed with incident and joined by **paper wipes**, native impact SFX and paper foley, landing on a face.

**The look is a specification, not a vibe.** Palette, exposure and pacing are numbers, and every failure is a number drifting off target. State targets as numbers in the prompt, then measure the output (`scripts/qc.py`). Adjectives don't survive the trip to the model; percentages do.

**The failure this skill exists to prevent** is not an ugly clip — it's a **boring** one: one slow scene per block, one camera drift, half the runtime doing nothing. It looks correct and feels dead. **Density is the deliverable.** The second failure, specific to story work, is a film that *narrates* instead of *showing*: if the voice is describing what the picture already says, the picture isn't working hard enough.

**The tool split (read this — it is the point of the pipeline):**

• **Everything generated runs on Unsora.** `create_image` (`nano-banana-pro`) for

the style key, the reference sheets and any hero frame → `create_video` (`seedance-2.0`) for every block → `create_voiceover` (ElevenLabs Eleven v3) for the narration → `create_music` (`mureka-7.5`) only if a bed is wanted. Each `create_*` is async: poll its matching `wait_for_*`.

• **Assembly is ffmpeg, locally.** Unsora has no assembler, so the cut, the

narration placement inside each block window, and the mix are ffmpeg commands you run yourself. This is an upgrade over any server-side assembler: block windows, VO centring and wordless blocks are all under your control instead of a server's.

Hard preconditions

• **Unsora MCP connected and authenticated.** Needs `create_image`,

`create_video`, `create_voiceover`, `list_voiceover_voices`, and the matching `wait_for_image` / `wait_for_video` / `wait_for_voiceover`; `create_music` + `wait_for_music` only if a score is requested. If they're missing, stop and say so — do not substitute another generator.

• **Local `ffmpeg` and `ffprobe`.** Not optional here — they *are* the assembler.

Without them you can generate clips but cannot deliver a film.

• **Web search** for the lore pass, whenever the story is a real tradition.
• **Recommended:** `python3` with `numpy`, `pillow`, `scikit-learn` plus `curl` for

the QC pass. If unavailable, run the eyeball checklist in `references/style-dna.md` and say QC was visual only.

Operating mode

Hands-off. The user delegates story choice, palette, script, voice, casting and assembly.

• Honour anything they specified — the myth, the tradition, the length, the

palette, the voice, the orientation. Decide everything else yourself.

• Right before the first **paid** generation, post one short plan message (the story

and which variant, palette, block count, the beat sheet in one line each, voice, rough credit estimate) so the user can interrupt — then **proceed immediately without waiting for approval**, unless they asked to be consulted. Note that the story shortlist below is a separate, earlier stop: by the time this plan message goes out, the myth is already settled.

• Never stop mid-pipeline to ask what a default answers. Delivering loose clips

instead of one assembled MP4 is a failure.

• **One exception: which myth.** If the user gave no story, present a shortlist and

wait for their pick (Phase S). Everything downstream — palette, voice, framing, casting, beats — you still decide. A topical explainer works with any trending subject, so picking one unasked is fine; a myth is the whole creative premise, and guessing it wastes the entire run. This is one stop at the front door, not a habit of asking.

Two style names collide, so be exact: this is the **three-colour cardstock** diorama, *not* the sepia-newsprint censor-bar diorama documentary. Do not use its letterpress prop text, and do not use any Mixed Media collage preset. This skill generates its own style key in Phase 1.

Defaults

| Setting | Default | Override when… | |---|---|---| | Style key | Generated in Phase 1 from the cardstock style-key prompt (`references/style-dna.md`), at the run's aspect ratio | user supplies their own reference image | | Palette | Auto-picked to fit the myth's world and mood (`references/palette.md`) | user names one, or the tradition has a strong colour identity | | Aspect | **16:9** — the concentric top-down hero shot nee