AbubakrChan/product-launch-motion / legacy
product-launch-motion-v1
VerifiedLegacy v1 (voiceover-led, GSAP + HyperFrames). Use when making, directing or rescuing a product launch video, promo, teaser, demo reel, sizzle, feature-announcement film or animated product explainer built from code — and when an existing one looks basic, reads as a slideshow, has reveals that miss the voiceover, an invisible cursor, an inaudible sound effect, a clipped or too-quiet mix, or every video coming out looking the same. Covers SaaS, apps, hardware, e-commerce, dev tools, services. Keywords include motion design, storyboard, voiceover sync, kinetic typography, GSAP, HyperFrames, Remotion, film grain, LUFS, ffmpeg.
references/11-creative-direction.md
View on GitHub# Creative direction
Owns Law 10: **no house style.**
Every other file in this skill is a discipline. A discipline applied identically to every
product produces identical films — which is the one failure mode that would make this whole
repo worse than useless, because it would mass-produce competent, interchangeable videos.
So read this file before you read the others, and read it again before you open an editor.
**The reference film is one point in the space, not the centre of it.** Its purple-to-orange
gradient, its two-register flip, its glass borders and its reconstructed inbox are answers to
one product's problem. Reaching for them by default means you skipped the only step that
makes a film feel made rather than generated.
## Contents
- [What is fixed and what is free](#what-is-fixed-and-what-is-free)
- [The direction dials](#the-direction-dials)
- [Deriving a direction from the product](#deriving-a-direction-from-the-product)
- [Diverge, then commit](#diverge-then-commit)
- [The signature move](#the-signature-move)
- [Five worked directions](#five-worked-directions)
- [Anti-sameness checks](#anti-sameness-checks)
- [Where not to be creative](#where-not-to-be-creative)
## What is fixed and what is free
| Fixed — the laws | Free — everything a viewer would actually describe |
|---|---|
| Truth: real numbers, real assets, real faces | Palette, and whether there is colour at all |
| Reveals land on their word | Ground: void, paper, material, photographic, field |
| Determinism: seek to T reproduces frame T | Typography and its personality |
| Rotation and scale on separate nodes | Camera: locked, dolly, orbit, continuous, none |
| The cursor is legible, if there is one | Whether there is a cursor, or any UI at all |
| Sound is measured, not guessed | Sound palette and whether music leads |
| Nothing is moving at a cut | Pacing, rhythm, shot lengths, cut style |
| One accent, with a written budget | Which accent, and what the budget is |
| Verified in the delivered file | Structure, register count, transition grammar |
| **No house style** | The signature move |
The test: **if two films built with this skill are indistinguishable, the left column was
applied and the right column was skipped.** The left column is craft. The right column is
the film.
## The direction dials
Pick a position on each dial *before* opening an editor, and write it into the storyboard.
This is a ten-minute exercise that decides everything downstream.
| Dial | ← one pole | other pole → |
|---|---|---|
| **Energy** | contemplative, long holds | relentless, cut on every beat |
| **Density** | one idea per frame, vast negative space | layered, maximal, many things true at once |
| **Ground** | void / flat paper | photographic, material, or a live field |
| **Depth** | resolutely flat 2D | deep 3D with real perspective |
| **Camera** | locked off, cuts do the work | continuous move across the whole film |
| **Type** | neutral grotesk, gets out of the way | editorial serif, mono, or expressive display |
| **Texture** | clinical, zero grain | film grain, halation, print, CRT, glitch |
| **Colour** | monochrome plus one signal | duotone, or the full brand spectrum |
| **Product** | reconstructed UI, literal | abstracted, metaphorical, or absent entirely |
| **Sound** | silence and UI ticks | score-led, or sound-design-led |
| **Voice** | narrated throughout | title cards only, no voice at all |
Two rules make this generative rather than decorative:
1. **Any two films you make should differ on at least four dials.** If they differ on one,
you made a variant, not a new film.
2. **Extremes are usually more interesting than centres.** A film that is *very* still or
*very* fast has a point of view. A film in the middle of every dial is the default look,
which is what everyone else's tool produces.
## Sameness hides in structure, not in surface
The dials above are necessary and **not sufficient**, and this is the most important
paragraph in the file.
When this was tested — two very different products, directed with no shared brief — the
surfaces came out genuinely unalike: one flat ledger-paper white with no camera and a
mechanical 120bpm foley grid, the other high-chroma riso silkscreen with continuous drift
and an analogue synth swell. By the dial test, a clear pass. Then the same director audited
its own pair and found **twelve structural convergences**:
- The same three-act skeleton, with the turn at the same proportional point
- **The same ending, beat for beat** — drain to the lightest value, shrink to a wordmark,
land one short line, in the final five seconds
- The same central rhetorical device (many collapsing into one)
- The same climax mechanic (one hero element swallowing the frame at the peak)
- The same copy placement (one quantified mid-film overlay, one end-card line)
- The same sound dramaturgy (sparse events over near-silence, then a sustained tone
arriving late to signal resolution)
- Both refusing a camera, both flat 2D, both with no human presence
- Both reaching into printed-matter history for their reference
- **Both rejecting their category's cliché in the same direction** — away from soft and
atmospheric, toward bold graphic flatness
That last one is the trap inside the trap: *deliberately breaking the category expectation
twice, the same way, is itself a house style.*
Two films can look nothing alike and still be the same film. So run the dials, then run
these — the structural axes, which is where the real defaults live:
| Structural axis | Vary it |
|---|---|
| **Rhetorical device** | many→one · before/after · a day in the life · testimony · pure demonstration · contradiction · absence (show the work that no longer happens) · a single object examined |
| **Arc** | problem→solve is one option. Others: start at the payoff and work back; a single unbroken demonstration with no argument; question held open until the last frame; chronological day |
| **Where the turn lands** | 40% is the default. Try 15% (cold open on the answer) or 75% (a long, patient problem) |
| **The ending** | Draining to a quiet lockup is the default *and everyone's default*. End loud. End mid-motion. End on the product still in use. End on the problem, absent |
| **The climax mechanic** | One hero element swallowing the frame is one option, not the option |
| **Opening state** | Empty ground that populates is the default. Start full and subtract. Start mid-action |
| **Copy placement** | If every film has one mid-film stat overlay and one end line, that is a template |
| **Human presence** | Most code-built films have none. A hand, a face, a voice, a real photograph is a bigger differentiator than any grade |
| **Sound dramaturgy** | Sparse-then-sustained-tone is the default resolution signal. Try: continuous bed that *stops*; no resolution at all; a single sound that recurs |
**The two-film test.** Before building, write your structure beside your last film's
structure. If the arc, the device, the climax and the ending match, you have made a variant
with a new palette — which is exactly what a viewer who has seen both will feel, even
though they could not name a single shared colour.
## Deriving a direction from the product
Four inputs. Work them in order; the fourth is the one that usually decides it.
**1 · The brand's own material.** Tokens, site, packaging, product photography. Take the
palette and the type seriously — but do not simply animate the website. A site has hover
states and scroll; a film has *time*, and time is a material the brand guidelines have
nothing to say about. Your job is to extend the brand into a medium it has never been in.
**2 · The category's visual expectation.** Name it explicitly in one sentence: "developer
tools are dark, mono, and terse." Then make a deliberate choice to **meet** it (fast
recognition, no friction — right when the product is unfamiliar and the category is trusted)
or **break** it (immediate differentiation — right when the product is a new answer in a
tired category). Both are valid. Doing it accidentally is not.
**3 · The audience's context.** Where does this play? A muted autoplay feed needs the claim
as type in the first two seconds and cannot rely on a voiceover. A conference screen can hold
a slow, quiet, wide shot. A sales-call attachment can be dense, because the viewer chose to
watch. The context changes the dials more than the product does.
**4 · The claim's emotional shape.** This is the real generator. Every claim has a feeling
under it, and the look should be that feeling expressed in a different medium:
| The claim is about | The film should feel |
|---|---|
| Relief from drudgery | Weight lifting: crowded → clean, dense → spacious, loud → quiet |
| Precision or trust | Instrument-like: grid, restraint, tabular figures, no flourish |
| Speed | Kinetic: hard cuts, speed ramps, motion blur, arrival before you expect it |
| Craft or taste | Slow, tactile, photographic, generous negative space |
| Power or scale | Wide, deep, three-dimensional, small human next to a big system |
| Play or delight | Overshoot, colour, sound-design-led, visual jokes that land in 200ms |
If your product's claim is relief and your film is relentless, the look is arguing against
the script. That mismatch is the most common reason a technically clean film feels wrong.
## Diverge, then commit
Do not write one direction and start building. Write **three**, deliberately different, then
kill two.
For each, one page, no images:
```
DIRECTION <name>
Thesis one sentence: what this film IS
Dials energy · density · ground · depth · camera · type · texture ·
colour · product · sound · voice
Palette ground, ink, accent — with the AA-safe small-text variant
Type family and the role it plays
Signature the one move only this film makes
Sound the palette, and whether music leads
Risk how this fails if executed badly
Wrong for the kind of product this would be wrong for
```
Then judge them against four questions:
1. **Does it serve the claim, or decorate it?**
2. **Does it look like this brand — but better than the brand currently looks?**
3. **Could a competitor use this direction unchanged?** If yes, kill it. It is a category
template, not a direction.
4. **Can it survive a bad day?** A direction that only works if every frame is perfect is a
direction you will abandon at 2am.
Then **commit and stop being creative.** Write the chosen direction into the storyboard's
direction block and execute it with total consistency. Invention happens at direction time;
execution is disciplined. A film that changes its mind halfway looks like two films.
## The signature move
Every film worth remembering does **one thing no other film does**. Not an effect — a move
that only makes sense for this product's claim.
It has to satisfy three constraints:
- **It expresses the claim.** If it would work equally well on an unrelated product, it is a
flourish, not a signature.
- **It repeats two or three times.** Once is an accident; four times is a tic. It should
appear early, pay off in the middle, and resolve at the end.
- **It is cheap enough to do perfectly.** An ambitious move executed at 70% is worse than a
simple one executed at 100%.
Ways to find one: take the product's core verb and make the film do it. If the product
*consolidates*, let scattered elements converge into every transition. If it *reveals*, let
the cursor be the only thing that uncovers type. If it *removes work*, let the frame get
literally emptier as the film runs. If it *connects*, make one continuous line the through-
element of every shot.
## Five worked directions
Five products, five deliberately unlike treatments. None of them is a preset — they are
here to show the *distance* between valid answers.
### 1 · Developer CLI — "Terminal as theatre"
Near-black ground, no gradient anywhere, monospace throughout, one phosphor accent. Camera
locked; every transition is a hard cut on a keystroke. Texture: faint scanline, no grain.
Sound: mechanical key travel, one success chime, otherwise silence.
**Signature:** the entire film is one continuously scrolling terminal buffer — every "scene"
is further down the same session, and the last frame scrolls back to the top.
**Wrong for:** anything sold to a non-technical buyer; the code-as-authority signal reads as
exclusion.
### 2 · Skincare DTC — "Macro and daylight"
Photographic ground, shallow depth of field, warm halation and real film grain. Editorial
serif, generous leading, almost no UI. Camera: slow push, never a cut inside a shot. Colour:
the product's own palette, desaturated one stop. Sound: room tone, water, no music bed.
**Signature:** every cut is a match-cut on the same circular form — a droplet, a jar lid, a
sun flare, a pupil.
**Wrong for:** a product whose claim is speed; this direction cannot be hurried.
### 3 · Fintech dashboard — "Precision instrument"
Swiss grid made visible, monochrome plus one signal colour used only for state changes.
Tabular mono for every figure. Zero texture, zero glass. Camera: orthographic pans that snap
to grid columns, no perspective at all. Sound: mechanical ticks on data changes, no score.
**Signature:** one continuous horizontal pan across a single enormous ledger, the whole film
reading left to right as one document.
**Wrong for:** a playful consumer product; this reads as institutional by construction.
### 4 · Consumer mobile app — "Thumb-first"
Vertical device in frame at all times, high energy, speed-ramped transitions, duotone
palette, chunky display type that overlaps the device. Sound-design-led with no narration —
title cards carry the words, because it will autoplay muted.
**Signature:** the phone never leaves the frame; the world rearranges around it between
scenes instead of the camera moving to a new place.
**Wrong for:** an enterprise buyer who needs to see the desktop reality.
### 5 · B2B SaaS platform — "Argument, then evidence"
This is the reference film, included so you can see it as *a* direction rather than *the*
direction. Two registers — dark for the argument, light for the product and its proof —
with each flip as a story turn. Kinetic type carries the argument; reconstructed UI carries
the demo. Camera restrained: one turn and one dolly per shot, never both prominent. Light
film grade. Narrated throughout.
**Signature:** one creator carried across four product shots, so three demos become one
continuous story.
**Wrong for:** a product with no UI to show, or an audience that will watch it muted.
## Do not inherit from the five above
The worked directions are there to show the **distance** between valid answers. They are
not a menu, and this is the failure mode they most invite: when a brief lands on top of one
of them — a skincare product, a fintech dashboard, a CLI tool — the honest first instinct
is to write that entry back out with new hex values. It will pass most checks, because it
is specific enough to feel like a choice.
So make the rule mechanical:
> **If your brief matches one of the five worked directions, that direction is
> automatically one of your three — and it is automatically one of the two you kill.**
Write it out, name what it inherited, and then build something that is *not* it. Its value
was showing you the obvious answer so you could recognise it and move past it. This applies
just as much to the reference film in `CASE-STUDY.md`: a B2B SaaS brief should
treat that film as the answer to beat, not the answer to reuse.
## Anti-sameness checks
Run these before you build. Each one has a right answer that is not "no".
- **Did you inherit from a worked direction in this file, or from the case study?** Name
the entry. Then kill it and write a third.
- **Are you using purple-to-orange?** Then you took the reference film's accent instead of
deriving one from the brand.
- **Does your film have glass borders *and* a gradient-mesh ground *and* specular sweeps?**
That is the default kit, applied wholesale. Pick the one that serves your direction and
drop the others.
- **Is your opening shot kinetic type on a dark ground?** It is the most common default in
the medium. Is it right for this claim, or just familiar?
- **Do you have a fake cursor in a film about a product with no UI?** Then you copied a shot
instead of choosing one.
- **Would this direction work unchanged for the product's closest competitor?** If yes, it is
a category template. Kill it.
- **Can you say the signature move in one sentence?** If not, you do not have one.
- **How many dials differ from the last film you made?** Fewer than four means you made a
variant.
- **Did you write three directions, or one?** One is not a choice.
## Where not to be creative
Invention belongs in what the film *is*. It does not belong in how it is built.
Do not get creative about: word-locked sync, determinism, the truth pass, the two-node rig,
sound measurement, the mastering targets, or verifying the delivered file. Those are not
stylistic positions — they are the difference between a film that works and a film that
ships broken. Every one of them exists because something failed in a measurable way.
**Be inventive about the film. Be boringly disciplined about the build.**
## Related
- `references/02-story-and-truth.md` — the claim your direction has to serve
- `references/09-shot-catalog.md` — constructions to adapt, never a running order to copy
- `references/06-look-and-grade.md` — the techniques, once you have chosen which you need
- `CASE-STUDY.md` — one direction, executed and revised seven times
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |