Image Maker Memory — Kontra Day

Overview

I generate UI design mockup images for the Kontra Day news-opinion site via the OpenRouter image API.

Get API key: source ~/.config/environment.d/10-secrets.conf (env var OPENROUTER_API_KEY)
Model used: openai/gpt-5-image-mini
Call format: curl .../chat/completions with {"model": ..., "messages":[{"role":"user","content":"Generate image: <detailed description>"}]}
Image returned as base64 in choices[0].message.images[0].image_url.url (data URI). Decode and save as PNG.

Output comes back as square 1024x1024 even when a wider aspect is requested. Don't expect strict size matching.
See skills/design-explore/SKILL.md for the full mockup-generating workflow and the Kontra Day design system.

Palette: canvas #fbf9f3 (warm newsprint off-white), wells #f0eee8/#eae8e2, hairlines #e4e2dd, ink #1b1c18 (soft charcoal, NOT pure black), muted #444748, faint #747878, vermilion #ad3222, dark inverse #1c1b1b + white text.
Typography: Newsreader (masthead/headlines), Literata (body 17-19px, lh ~1.6), Work Sans (UI labels/caps).
Shape: 0px radius everywhere, flat + hairline rules, no shadows/gradients.
Spacing: 8px grid; content column max 42rem centered.