Auto-banner / brand-image skill (Canva replacement)

A skill that generates banners, story plates, podcast covers, and YouTube thumbnails in a saved brand style. The style definition lives as a markdown file the agent reads ({rule} shaper-style.md): brand colors, typography, layout rules, do/don't examples. Replaces a constellation of custom GPTs and a Canva seat per marketer.

Routine in-brand assets stop being a designer-blocking task. The brand spec lives where the team can edit it (markdown in the vault) rather than locked in a SaaS account.

1. Write {rule} shaper-style.md in the vault: brand colors with hex codes, typography stack, layout rules, do/don't examples, image-gen prompts that produce on-brand backgrounds.
2. Build skill (e.g., banner-3-to-1, cover-generator) that reads the style file plus a content topic.
3. Skill decides when to call the image-gen model for backgrounds vs reusing brand asset library.
4. Expose a small web UI for non-engineers: text input, color/background toggles, "regenerate" button.
5. Also expose a terminal mode for power users who want to iterate via voice.
6. Output formats per channel (3:1 for Telegram cover, 1:1 for Instagram, etc.) baked into named skill variants.

• Keep a designer in the loop for two-pixel adjustments. The skill solves the 80% of routine assets; it does not replace a designer for hero brand work.

• Multi-tool brand-style stack. Beyond Canva-replacement, a more comprehensive designer stack pairs (a) a vision model with brand-style ingestion (see [new-N14-15]) for on-brand asset generation, (b) Midjourney or equivalent for hero imagery, (c) a local LLM where data residency regulation requires it (e.g. for kid-content under regional residency rules), (d) ChatGPT for copy. Same operator orchestrates all four; the skill router picks the right tool by asset class.