relume.ai

# Start here: relume.ai

Observed 2026-10-28T00:45:20.671Z; default light theme. This is a compact handoff, not the full evidence archive: everything you need to start is here, the detailed files are one step away.

## How to use this pack

1. Read `START_HERE.md` and `tokens.css` first: the priority values. Keep observed values; a value marked inferred or fallback is a suggestion to resolve against the screenshots.
2. Open `screenshots/desktop-1440-fold.png` and `screenshots/phone-390-fold.png` and match the composition, density, whitespace and image treatment; the values do not carry those. Treat the inside of an image or canvas as a visual inference, not measured DOM.
3. Read `FONTS.md` and load `fonts.css` before `tokens.css`. Keep exact weights and variable-font settings and verify the fonts that actually render. If a font is unavailable, name the substitute and compare its width and apparent weight before touching the type scale.
4. Before drawing any logo or icon, read `BRAND.md` and open `screenshots/brand-logo.png`, `screenshots/brand-icon-1.png`, `screenshots/brand-icon-2.png`: for a new product, create an original mark with the observed proportions and treatment; never guess a shape from a label.
5. Before building any control, read `COMPONENTS.md` (every variant and measured state, with a crop of each), use `components.css` for the CSS and `demos/` for a live copy of each component. A component listed as not found does not exist on the site; do not invent it.
6. Read `REGIONS.md` and inspect its context, detail and icon crops: reconstruct outer and inner surfaces, per-layer gradients and shadows and group spacing (`regions.css` has the recipes), then check the whole page again.
7. Read `BEHAVIOR.md` for hover, focus, tooltips, menus, motion, scrolling, tables, forms, text and each breakpoint, and reproduce the behaviour, not only the colours. Build both themes; `DESIGN.md` pairs the colours.
8. `layout.html` is a generic page composed only from this pack's measured values, with placeholder copy (`layout-alt.html` in the other theme): open it to see the colours, type, recipes and spacing working together and to check your own wiring of the fonts and tokens against it. It is not the site's composition; the screenshots are.
9. `DESIGN.md` has the whole system when a value is missing above. `evidence.json` is diagnostic data: never in the first read.
10. When done, render at desktop, 390px and 320px, compare with the screenshots, fix the biggest visible mismatch first (usually type size, whitespace or image treatment), recheck keyboard focus, overflow, navigation and forms, and repeat.
11. For another site's design, run `npx stylepack-cli <host>` in the project (it downloads that pack and records it the same way), or fetch https://stylepack.ai/packs/<host>/pack.zip; one question at a time is at https://stylepack.ai/packs/<host>/fonts, /colors and /buttons.

## Priority values and evidence

Values labeled fallback were not observed for that role; inferred-role means a visible heading was mapped to a display role. Keep observed values, but treat fallbacks as suggestions.
Keep measured sizes, tracking, radii and colors unless the new content requires an explicit adaptation. Do not enlarge the headline simply to fill a different line length.
Content width: 1280px. Inferred spacing base: 4px; sampled scale: 4, 8, 12, 16, 20, 24, 32, 72, 80px. Observed CSS breakpoints: 479, 700, 767, 768, 991px.
Colors: background=#f1f0ee; surface=#ffffff; surface-2=#e4e3e1; foreground=#101010; muted-foreground=#343434; primary=#ff00f4; on-primary=#111111; secondary=#98eafd; border=#dcdbd9.
- primary: Selected from measured button, link, text, painted-area and theme colors. It is not proof of a filled primary button; follow the observed component recipes.
Radii: button 6px, card 0px, input 0px, image 0px.
Shadows: rgba(27, 25, 25, 0.1) 0px 0px 0px 1px inset; rgba(27, 25, 25, 0.1) 0px -1px 0px 0px inset; rgb(27, 25, 25) 0px 0px 0px 1.25px inset.

## Typography

Load fonts.css before applying tokens.css. FONTS.md contains the observed face/weight inventory and rendered-font checks. A CSS family name alone does not load a font.
Copy the supplied @font-face sources verbatim; do not replace a captured face with Google Fonts merely because its name looks similar. Keep fonts.css linked (or inline it verbatim for a single HTML deliverable). It includes measured semantic font defaults; tokens.css supplies scale and spacing. Use each role as a complete recipe: family, exact numeric weight, size, line height, tracking, font-feature-settings (including cv/ss stylistic sets), optical sizing and variable-font settings. A font shorthand resets feature settings; reapply the supplied longhands after it. Do not round 510 to 500/600, or turn a measured light style into a bold heading.
Use the captured family first. Self-hosted is a hosting location, not an availability verdict. An unverified probe is not evidence of failure; do not choose the optional fallback merely because it is familiar. Substitute only after an actual failed destination request or missing licensed source; record that reason and the replacement. A transient network error merits a retry before redesigning the typography.
Fallback roles are unmeasured suggestions. If a role is absent, start from the nearest observed text treatment; do not assume headings must be bold.
- Geist Variable: observed weights 400, 500; source loading recipe in fonts.css.
- Geist Variable delivery by face: confirmed 100 900 normal; failed none; unverified none. Confirmation is from one isolated-origin check, not every deployment. Keep confirmed faces; attempt unverified faces. A failed unused face does not invalidate a working regular/bold face.
After loading, inspect the browser's Rendered Fonts for the headline, body and button. Computed font-family/font-weight and document.fonts.check() alone do not prove which face drew the glyphs.

- headline-display: Geist Variable; 56px/67.2px; weight 400; tracking -1.68px; observed, source h1.
- headline-lg: Geist Variable; 40px/48px; weight 400; tracking -0.8px; observed, source h2.
- headline-md: Geist Variable; 18px/25.2px; weight 400; tracking -0.36px; observed, source h3.
- body-md: Geist Variable; 18px/27px; weight 400; tracking normal; observed, source body.
- label-md: Geist Variable; 12px/18px; weight 400; tracking normal; observed, source label.
- button: Geist Variable; 14px/20.02px; weight 500; tracking normal; observed, source button.
- Font Geist Variable: self-hosted; use its captured loader first. Self-hosted does not imply unavailable or licensed for reuse; substitution is conditional, not a second design choice.

## Regional composition

Read REGIONS.md and open the referenced context, detail and icon crops before drawing complex previews. Build outer shell, inset content panel and internal groups as distinct layers where observed. Preserve which surface owns each gradient, border and shadow; do not apply one generic card treatment to every layer.
For images, canvas, video and external SVG, use the rendered crops to infer visible internal groupings. Keep those inferences separate from measured DOM facts. Do not treat an illustration as a functional app or paste the screenshot as the finished UI. Return to the whole-page screenshots to match scale and placement.
- r1: 909×605px at page (392, 1833); 24 measured layers, 12 icon samples, 3 opaque media layers. screenshots/region-r1-context.jpg, screenshots/region-r1-detail.png
- r2: 1280×868.3px at page (80, 3325.3); 24 measured layers, 4 icon samples, 8 opaque media layers. screenshots/region-r2-context.jpg, screenshots/region-r2-detail.png, screenshots/region-r2-icons.png
- r3: 1280×833.6px at page (80, 5774.5); 22 measured layers, 12 icon samples, 3 opaque media layers. screenshots/region-r3-context.jpg, screenshots/region-r3-detail.png, screenshots/region-r3-icons.png

## Component recipes

- primary button: bg #ff00f4; text #ffffff; border none; radius 6px; height 32px; padding 6px 12px; type 14px/400; shadow none.
- secondary button: bg #ffffffeb; text #101010; border none; radius 9999px; height 88px; padding 0px 0px; type 14px/400; shadow rgba(0, 0, 0, 0.24) 0px 8px 32px 0px.
- ghost button: bg transparent; text #161616; border none; radius 6px; height 25px; padding 0px 8px; type 14px/400; shadow none.
- Card: {"bg":"#ffffffcc","border":"1px solid #1b19191a","radius":"16px","padding":"0px","shadow":"none","w":640,"h":174}.

## Responsive and interaction evidence

- desktop 1440px: content 1280px, gutter 432px, H1 56px, 2 lines, center, header 56px, 4 links, grids 2→1 col; 2→2 col; 2→1 col.
- laptop 1024px: content 922px, gutter 192px, H1 56px, 2 lines, center, header 56px, 4 links, grids 2→1 col; 2→1 col; 2→1 col.
- tablet 768px: content 691px, gutter 64px, H1 52px, 2 lines, center, header 56px, hamburger (Open menu), 1 CTA, grids 2→1 col; 2→1 col; 2→1 col.
- phone 390px: content 351px, gutter 20px, H1 40px, 3 lines, center, header 56px, hamburger (Open menu), 1 CTA, grids 2→1 col; 2→1 col; 2→1 col.
H1 goes 56px → 40px (71%), 2 → 3 lines.
Nav collapses to a hamburger by 390px (first seen at 768px); header 56px → 56px.
Card grids: 3 columns → 1 column on phones (gap 62.26px → 77.84px).
Body text is 14px at the sampled desktop and phone widths.
- Hover button-primary: moves down 1px.
- Hover button-secondary: no visible change.
- Hover nav-link: background transparent → #1616160f.
- Focus button-primary: outline 1px auto rgb(16, 16, 16).
- Focus input: no visible focus style measured (may rely on :focus-visible with keyboard only).
514 elements carry transitions; typical 0.2s with ease, mostly on opacity, color, transform. 10 running keyframe animations (ra-caret-blink 1.06s infinite; ra-trail-run 5s infinite; logoloop 30s infinite).
Alternate dark theme observed; use DESIGN.md for paired colors.

## Logo and icons

Logo lockup: 87×23px, external image or text; inspect crop. See BRAND.md and its crops for shape evidence.
Icon: 12×12px; filled; stroke none observed, caps not applicable.
Icon: 30×30px; filled; stroke none observed, caps not applicable.

## First images to open

- screenshots/desktop-1440-fold.png: Desktop 1440, above the fold
- screenshots/brand-logo.png: primary logo lockup
- screenshots/phone-390-fold.png: Phone 390, above the fold

## Verify the build

Render at desktop, 390px and 320px. Inspect the result, compare against the supplied references, fix the largest visible mismatch, and recheck keyboard focus, overflow, navigation and form behavior. Match the visual language while following the requested product content.