Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
45 changes: 45 additions & 0 deletions skills/hyperframes/references/prompt-expansion.md
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,51 @@ The quality gap between a single-pass composition and a multi-scene-pipeline com

**Do not skip. Do not pass through.** Single-scene compositions and trivial edits are the only exceptions.

## Before you enrich: calibrate to content culture

Enrichment without calibration is the main failure mode of always-enrich. Before picking decoratives, motion, and narrative arc, identify the composition's **genre** and pick the energy register the genre expects. Reach for enrichment inside that register.

The failure mode this rule corrects: a running-shoes teaser treated as an introspective narrative (stillness → ignition → claim) is clever but wrong — consumer-brand launches exist to pump the viewer up, not to make them contemplate. Similarly, a documentary about a 1936 photograph treated with tech-product chrome (ghost watermarks, coord stamps, corner registration ticks) fights the tone.

| Genre | Energy register | What the genre wants | What the genre does NOT want |
| --------------------------------------------------------------- | -------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| **Consumer brand launch** (sports, fashion, beverage, CPG) | Bold, kinetic, in-your-face | Display typography at scale (200px+), marquee motion, beat-synced cuts, oversized color blocks, counters ticking live, confidence in short bursts | Introspective narrative arcs, contemplative opening scenes, whisper captions, "journey" framing |
| **Product demo** (SaaS, B2B, developer tools) | Confident, measured, purposeful | The product doing its actual thing — pipeline moving, values changing, workflows running. Arc follows the product's real flow, not a marketing template | Problem-hook → stat → product → proof → CTA when the product itself is more interesting than the pitch |
| **Technical explainer** (science, engineering, how-things-work) | Clear, labeled, progressively revealed | Diagrams that build — each scene adds a labeled entity to the previous one. Restrained. Numbers with units. | Dramatic reveals, bombastic typography, "sci-fi" moves |
| **Documentary / archival** | Quiet, reverent, observational | Long holds (4+ seconds of stillness per scene), restrained typography, the subject doing the work. Minimal decoration. | Tech-product chrome (registration marks / coord stamps / ghost words / tick rails), bouncy eases, ambient "breathing" on everything |
| **Editorial / brand story** (luxury, fashion, lifestyle) | Warm, writerly, mood-driven | Serif headlines, pull-quotes held unchanged for seconds, generous whitespace, mono metadata only for factual anchors (dates, places, credits). Magazine feel. | Dashboard chrome, data-UI patterns, tabular numbers, dense HUDs |
| **News / social / viral** | Snappy, high-contrast, caption-forward | Big readable text, fast cuts, stacked caption layers, short scene durations (1.5–3s), urgency | Slow eases, ambient drift, "breathing" decoratives |

**Always-enrich has a known over-rotation: it reaches for narrative reframes on any prompt.** For brand/consumer content, that reframe costs you the genre's native energy. The expansion should choose the register FIRST, then decide what to enrich.

### How to apply

1. Read the user's prompt. Identify the single-most-likely genre from the table above.
2. Pick the energy register. Note what it does NOT want.
3. NOW enrich — decoratives, motion, narrative arc — staying inside that register.

If the prompt's genre is ambiguous, check `design.md`'s mood field — it often disambiguates (`mood: bold, athletic, kinetic` means brand-launch register, not documentary).

## Use the real subject when it exists

If the composition is about a specific, named, real-world artifact — a photograph, painting, song, public figure's recorded words, a real company's UI, a historical event — and that artifact is accessible via public sources, **use it**. Don't abstract it to a placeholder to preserve a skill rule.

- A documentary about Dorothea Lange's 1936 "Migrant Mother" → fetch the photograph (it's in Library of Congress FSA/OWI collection, public-domain-equivalent, IIIF accessible). Don't ship a `[ PHOTOGRAPH ]` placeholder caption.
- A piece about Vermeer's "Girl with a Pearl Earring" → fetch the painting (Mauritshuis has open-access imagery).
- A 1969 moon landing piece → specific NASA photographs exist in NASA's open archive.
- A piece about a specific public figure's speech → quote their actual words when available.
- A product demo for a well-known company → their actual UI and brand are canonical; use them unless the prompt says otherwise.

**R4 (persistent-element continuity morph) applies to HOW you present the subject**, not WHETHER to use a real one. When both are available, use both — fetch the real artifact AND structure it via a persistent overlay that migrates across transitions.

### When to abstract instead

- The subject doesn't exist yet (a fictional brand launch).
- The subject is private / proprietary / unavailable (an unreleased product).
- Using the real subject would require violating license or consent (a living person who hasn't consented to this use).

These are the exceptions. The default is: use what's real.

## What to generate

Expand into a full production prompt with these sections:
Expand Down
Loading