Logo Pack

Turn a brand into a production-ready logo pack — a tailored, professional set of logo directions and the prompts to generate them. The pack is diverse on purpose (several styles) but every option is on-brand and specific to this product, never generic clip-art.

It is brand-aware: it reads docs/brand.json (the same brand-as-data file the rest of the ship pack uses) and checks for an existing logo before doing anything.

The first question: refine or create?

Always start by scanning what already exists:

node scripts/scan-brand.mjs        # or: --json

It reports the brand.json, any brand guide, and any existing logo files, then recommends:

• A logo already exists → ANALYZE & REFINE. Read the existing mark (look at the actual image), extract its visual DNA — forms, weight, color, personality — and produce variations that stay true to it: a one-color version, a monochrome version, a simplified app-icon, a cleaner refinement. Don't throw away brand equity unless the user explicitly asks to redesign.
• No logo (or the user says "make a new one") → CREATE. Design fresh directions from the brief. Still diverse-but-professional, still on-brand.

State which mode you're in and why.

Build the brief

Before writing prompts, capture a short brief (save it to docs/brand/logo-brief.md):

• Name + what it is (one line) + who it's for
• Personality — 3–5 adjectives (from the tagline, brand, or ask the user)
• Colors — from brand.json (brand, accent, plus neutrals)
• Must-work conditions — small sizes, one color, dark and light backgrounds, app icon
• Avoid — clichés that don't fit (generic globes, swooshes, gears, gradients-for-no-reason)

If the product's purpose isn't clear from brand.json/context, ask the user one line about it.

Write the prompt set

Produce a tailored prompt per style, following references/prompt-styles.md. Always include these four (the core set), then add supporting styles as useful:

Each prompt is tailored to this product (its purpose and audience drive the imagery), names the brand colors, and specifies logo-grade output: flat/vector-like, high contrast, legible at 16px, transparent background, no stray text. Save all of them to docs/brand/logo-prompts.json.

Generate (production step, generator-agnostic)

The skill writes prompts; you choose the engine. Hand logo-prompts.json to whichever is set up:

• using-local-ideogram4 skill — best for wordmarks/text (Ideogram renders type well).
• gemini-imagegen skill (Nano Banana Pro) — strong for clean marks + quick variations.
• Ad Lab — node D:/git/mkm/ad-lab/scripts/generate-image.js -p "<prompt>" -o docs/brand/logos/<id>.png.
• Replicate (in-app) — model google/nano-banana-2.

Save outputs to docs/brand/logos/<style-id>.png. Generate 2–4 per prompt; logos are a numbers game — make many, keep few.

Review and iterate

Lay the candidates out (a simple contact sheet or just the folder), pick the strongest, then refine that one with a follow-up prompt (tighten spacing, true the geometry, lock the color). For the winner, export the standard set: full-color, one-color, monochrome, and a square app icon / favicon. Update brand.json if the logo establishes or shifts the brand colors.

Quality bar

• Specific, not generic. The mark should only make sense for this product. If the same logo would fit any company, the prompt is too vague.
• Works in one color and at 16px. If it dies when shrunk or flattened, it's not a logo.
• On-brand. Honor brand.json colors and the personality. Never use purple for the user's apps unless their brand says so.
• Honest. Prompts are starting points — generation always needs a human pick + a refine pass.