Framefields Styles

---
name: framefields-styles
description: Choose and combine the creative direction of a framefields video — its look (Motion, Collage, Blueprint, Keynote …), format (widescreen, vertical 9:16, square, 15/30 s ad), genre (explainer, product launch, brand film, lyric video) and media (code-drawn, 3D, footage, AI-generated through the framefields gateway) — or start from a preset. Use when someone asks for a video "in the style of" something, names a platform or length, asks what kind of video to make, wants AI-generated clips to match drawn graphics, or before choosing palette, type, pacing or music for a new film.
---

# framefields-styles

A video's direction is four parts, each a short file in this skill:

- **Look**: what it looks like (palette, type, texture, motion character, grade, and art direction for generated media).
- **Format**: where it plays (size, fps, length, safe areas, cut rhythm).
- **Genre**: what kind of video it is (the beats, what the audio does, what's on screen).
- **Medium**: where the pixels come from (code-drawn, 3D, footage, AI-generated). A film can use several. Media are decided per piece: by the user when they say, otherwise by you from the request.

A **preset** names a good combination of a look, a format and a genre.

## Choosing

1. If the user named a preset or parts, use them.
2. Otherwise, infer the format (from the platform or length they mention) and the genre (from what the video is for), then offer two or three looks that pair well, by name and one line each. Ask only what you can't infer.
3. Pick the media from what the request needs (say which), unless the user named them.
4. Read only the files you use: one look, one format, one genre, and each medium.

## Combining

Each part writes only in its own area, so they combine without overriding each other. When two pull against each other:

- **Format** wins on the frame: size, safe areas, length, how often to cut.
- **Genre** wins on structure: the order of the beats, what the audio does.
- **Look** wins on everything visual: color, type, texture, easing, grade.
- **Medium** wins on production: tools, scripts, cost, verification.

A slow look in a short format keeps its visuals with fewer elements; it does not slow the format down. `clashes` in a part's frontmatter mark pairs to handle with care, never pairs to refuse.

From code, `framefields/styles` has the same library: `combineStyles({ look, format, genre, media? })` returns the brief these rules produce, `styleClashes(mix)` the pairs to watch, and `styleLibrary` every part.

## Generated media matches the look

When a medium uses AI generation, it generates images (video only if the user asks), and every prompt ends with the look's `ai.direction` and `ai.avoid`, word for word; the engine then moves the stills the way the look's `ai.motion` says. The look's grade then goes over generated and drawn material alike (`comp.apply`). That shared text and grade are what make generated images and code-drawn type read as one film. The steps are in [media/ai-generated.md](media/ai-generated.md).

## The library

<!-- style-library:start (generated by pnpm run sync:style-library) -->

### Looks

| Look | What it is | Tags |
|---|---|---|
| [Blueprint](looks/blueprint.md) | Diagrams that build step by step on a quiet grid, with labels, arrows and code. | technical, diagram, clean |
| [Collage](looks/collage.md) | Paper collage, highlighted documents and slow pushes on archival images. | editorial, texture, documentary |
| [Keynote](looks/keynote.md) | A product on black, slow light across it, one line of type at a time. | product, premium, cinematic, 3d |
| [Motion](looks/motion.md) | Kinetic type, bold flat shapes and snappy easing, cut to the beat. | type, graphic, music |
| [Risograph](looks/risograph.md) | Two-ink zine printing; misregistered blue and pink, grain and big cut-out shapes. | print, texture, graphic, playful |
| [Swiss](looks/swiss.md) | The International Typographic Style; a strict grid, flush-left type and one red. | type, graphic, grid, clean |

### Formats

| Format | Frame | Length | What it is |
|---|---|---|---|
| [Ad spot](formats/ad-spot.md) | 1920×1080, 30 fps | 15–30 s | A 15 or 30 second spot with a fixed shape; hook, one message, brand, call to action. |
| [Bumper 6s](formats/bumper.md) | 1920×1080, 30 fps | 6–6 s | A six-second unskippable spot: one image, one line, the name. |
| [Cinema](formats/cinema.md) | 1920×1080, 24 fps | 30–180 s | 2.39:1 widescreen with letterbox bars: trailers, title sequences and film-like pieces. |
| [Square](formats/square.md) | 1080×1080, 30 fps | 6–60 s | 1:1 for feeds and ads that play muted; centered and self-explaining. |
| [Vertical](formats/vertical.md) | 1080×1920, 30 fps | 10–60 s | 9:16 for Reels, Shorts and TikTok; a hook first, big captions, quick cuts. |
| [Widescreen](formats/widescreen.md) | 1920×1080, 30 fps | 30–600 s | 16:9 for YouTube, sites and presentations; room for a story to breathe. |

### Genres

| Genre | Beats | Audio |
|---|---|---|
| [Brand film](genres/brand-film.md) | Open → Build → Peak → Resolve → Name | music, sound |
| [Data story](genres/data-story.md) | Claim → Context → Build → Turn → Takeaway | voice, music |
| [Explainer](genres/explainer.md) | Question → Context → Steps → Payoff → Recap | voice, music |
| [Lyric video](genres/lyric-video.md) | Intro → Verse → Chorus → Bridge → Outro | lyrics, music |
| [Product launch](genres/product-launch.md) | Tease → Reveal → Features → Name → Call to action | music, sound |
| [UI demo](genres/ui-demo.md) | Problem → Open → Do → Result → Next | voice, music, sound |

### Media

| Medium | What it is | Uses credit |
|---|---|---|
| [3D](media/3d.md) | Real 3D scenes: products, objects and type with cameras, lights and materials. | no |
| [AI-generated](media/ai-generated.md) | Images, music and voice made through the framefields AI gateway, matched to the look and moved by the engine. | yes |
| [Code-drawn](media/code-drawn.md) | Everything drawn by the engine: type, shapes, charts, components. Free, exact and editable. | no |
| [Footage](media/footage.md) | The user's own clips and photos, cut, graded and annotated, with on-device vision. | no |
| [Mock-up → AI edit](media/mockup-edit.md) | Build the shot in the engine for exact layout, then let an image model make it real. | yes |

### Presets

| Preset | Look | Format | Genre |
|---|---|---|---|
| [Collage explainer](presets/collage-explainer.md) | collage | widescreen | explainer |
| [Swiss data story](presets/data-story.md) | swiss | widescreen | data-story |
| [Launch teaser](presets/launch-teaser.md) | keynote | ad-spot | product-launch |
| [Motion](presets/motion.md) | motion | widescreen | brand-film |
| [Social vertical](presets/social-vertical.md) | motion | vertical | brand-film |
| [Technical explainer](presets/technical-explainer.md) | blueprint | widescreen | explainer |
| [Zine bumper](presets/zine-bumper.md) | risograph | bumper | brand-film |

<!-- style-library:end -->

## Adding a part

One markdown file in its kind's folder, named by its id. Frontmatter holds the facts (see an existing file of the same kind); the body has only that kind's `##` sections, in order:

- look: Look, Type, Motion, Grade?, Build it with, Avoid?
- format: Frame, Timing, Safe areas?, Build it with?
- genre: Beats, Audio, On screen, Build it with?
- medium: Pipeline, Cost?, Consistency?, Verify, Build it with?
- preset: Note? (only)

Names in backticks under "Build it with" must be real `framefields` exports or `components/<id>` registry items; the tests check them. Run `pnpm run sync:style-library` after any change.