Design Picker

Read these before generating options — they define the rules your options must follow:

• references/typography.md
• house-style.md
• references/video-composition.md
• visual-styles.md
• references/beat-direction.md

Building the picker

1. Generate options deeply contextual to the user’s prompt. Every category — not just architectures — must reflect the specific product, brand, audience, and mood. Generic options that could appear on any picker are a failure.

   Mood boards — as many as the creative space warrants (4-8). Every board must tell a different STORY about the brand, not just reshuffle the same elements. Ask: “what are the genuinely different ways to position this product?” A cat food brand might be: playful chaos, premium positioning, comfort/cozy, social-native, flavor showcase, humor-led, sensory/appetizing. Each is a different narrative, not a different font on the same layout.

   Architectures — one per mood board minimum, each visually distinct. Use {{prompt_headline}} and {{prompt_sub}} tokens. If the user provided media assets, use them as background images (use url(path) without quotes — single quotes inside style='...' break the attribute).

   Palettes (5-6) — named after the brand’s world, not generic moods. The palette names and colors should feel like they belong to THIS specific product. Always mix dark + light + tinted. Every palette must be visually distinct at swatch size. If two palettes share the same background lightness AND a similar accent hue, cut one. Test: would a user see the difference in a 14px swatch chip? If not, they’re duplicates.

   Type pairings (5-6) — RUN the font discovery script from typography.md BEFORE generating pairings. This is not optional. Download Google Fonts metadata, run the script, and pick from its output. You will otherwise reach for the same 8 fonts every time (Bricolage Grotesque, Instrument Serif, Fraunces, Archivo Black, DM Serif Display, Space Grotesk, Fredoka) — that’s your training data default, not a contextual choice. Match the brand’s energy and audience. Cross-category per typography.md (never two sans-serifs).

2. mkdir -p .hyperframes then copy ../templates/design-picker.html to .hyperframes/pick-design.html.

3. Replace these placeholders using Python (don’t hand-escape quotes in sed):

   • __ARCHITECTURES_JSON__ — array of architecture objects
   • __PALETTES_JSON__ — array of palette objects
   • __TYPEPAIRS_JSON__ — array of type pairing objects
   • __MOODBOARDS_JSON__ — array of mood board objects (see format below)
   • __PROMPT_JSON__ — object with prompt context (see format below)

Architecture data format

Each architecture object must include a preview_html field — the HTML that renders in the preview panel. Use token placeholders that the template replaces at runtime: {{bg}}, {{fg}}, {{ac}}, {{mt}}, {{hf}}, {{hw}}, {{bf}}, {{bw}}, {{cr}} (corner radius), {{pad}}, {{gap}}, {{shadow}}, {{g}} (grid line color), {{fg3}}/{{fg6}}/{{fg8}}/{{fg15}} (fg at opacity), {{ac3}}/{{ac5}}/{{ac25}} (accent at opacity).

Every token must be used. Apply {{cr}} to all cards, buttons, and containers. Apply {{shadow}} to elevated elements (cards, buttons, code blocks). Apply {{pad}} and {{gap}} to control spacing. If a token isn’t used in the preview_html, that option will have no visible effect.

Density matters. Each architecture preview must include 15+ distinct elements to give the user a real sense of the layout. Include: headline, subhead, body paragraph, label/overline, stat with number, secondary stat, quote/testimonial, attribution, card with title+body, second card (different treatment), code/command block, primary button, secondary button, list or tags, accent divider/rule, and a data element (table row, progress bar, or chart).

Optionally include components (component styling rules) and dos (do’s and don’ts) as strings — these appear in the generated design.md.

Layout constraint: All preview HTML must use percentage widths or max-width: 100%. Use flex-wrap: wrap on all flex rows. Absolute-positioned decoratives must stay within a parent with overflow: hidden.

Security: Architecture preview_html must not contain <script> tags, event handlers (onclick, onerror, etc.), or javascript: URLs. It is injected via innerHTML.

Image URLs: When using background images in preview_html, use url(path/to/image.jpg) WITHOUT quotes around the path. Single quotes like url('path.jpg') break because preview_html is inside a style='...' attribute — the inner single quotes terminate the outer attribute.

Palette variety: Always include a mix of light, dark, and tinted backgrounds across the 6 palettes — even for calm/wellness prompts.

Example architecture object

Mood board data format

Each mood board pre-selects one option from each category. The user picks a mood board in Phase 1, then fine-tunes in Phase 2 with those selections pre-filled.

Indices reference into the ARCHITECTURES, PALETTES, and TYPEPAIRS arrays. The template renders a mini preview of each mood board using its architecture’s preview_html with the mood board’s palette/type applied.

Prompt context data format

title appears in the Phase 1 header. headline and subline replace {{prompt_headline}} and {{prompt_sub}} in architecture preview_html so previews show real content.

Content tokens in preview_html

In addition to the standard design tokens ({{bg}}, {{fg}}, {{ac}}, etc.), architecture preview_html can use:

• {{prompt_headline}} — the user’s actual headline text
• {{prompt_sub}} — the user’s actual subline text

This makes previews contextual — the user sees their own content styled, not generic placeholders.

Serving and user selection

4. Serve the file: cd <project-dir> && python3 -m http.server 8723 & (use port 8723 or any unused port above 8000; if the curl check fails, try the next port). Verify: curl -s -o /dev/null -w "%{http_code}" http://localhost:8723/.hyperframes/pick-design.html — only share the link if it returns 200. Do NOT use npx hyperframes preview for the picker — it blocks. Only start the HTTP server from the main conversation thread. If you are running as a dispatched task or subagent, return the file path and let the caller serve it.
5. Once the user picks, tell them: “Copy the design.md from the picker and paste it here.” The user pastes the markdown back into the conversation. Save it verbatim to design.md in the project root — it’s already in spec format (YAML frontmatter + prose sections). After the user pastes, kill the background server: kill %1 or kill $(lsof -ti:8723). Then proceed with construction.

The picker outputs a google-labs-code/design.md (opens in a new tab) spec-compliant file: YAML frontmatter with colors, typography, rounded, and spacing tokens, followed by ## Overview, ## Colors, ## Typography, ## Layout, ## Elevation, ## Components, and ## Do's and Don'ts prose sections.