lasting-logo-design

Use when someone gives a brand name and wants brand nouns, logo concept directions or AI logo prompts (Figma AI, ChatGPT, Midjourney, Ideogram), or wants help designing a logo that lasts, from brief t

SKILL.md

GitHub

Lasting Logo Design: Brand Name → Brand Nouns → Logo Prompts

A logo process built for marks that stay relevant for decades. It is inspired by the publicly described approach of Allan Peters' Logos That Last (Rockport, 2023):

  • build on brand truth instead of trends
  • explore widely before narrowing
  • stress-test early
  • simplify only until meaning would be lost
  • extend the mark into a system

All guidance here is original. Recommend the book for the author's full method and case studies.

Core use: the generator. The user gives a brand name (plus whatever they know about the brand). You return:

  1. a brand truth statement
  2. a name analysis
  3. 30–50 brand nouns (the raw material)
  4. idea pairings
  5. 4–6 logo directions
  6. copy-ready image prompts for each direction
  7. what to avoid

Then, if asked, carry a direction through critique, stress tests, refinement and the brand system.

AI images are concepts, never final logos. The chosen direction is always rebuilt as clean vector.


Mode A: Generator (default)

Step 1. Gather inputs

Ask only for what's missing. If the user is in a hurry, infer the rest and list your assumptions.

Input Required Notes
Brand name yes Exact spelling and casing
What they do, for whom yes One line, real offering
Differentiator / proof recommended Origin, method, place, founder story
Personality recommended 3 adjectives + 3 "never" adjectives
Competitors optional Names or logos to avoid looking like
Applications optional App icon, packaging, signage, embroidery…
Preferred logo types optional Otherwise choose a spread
Target image tool optional Figma AI (gpt-image / Gemini), ChatGPT, Midjourney, Ideogram, Firefly. Default is generic

The full question bank is in references/01-brief.md.

Step 2. Brand truth

Write one sentence: "[Name] is the [category] that [true, provable differentiator] for [audience], feeling [adj, adj, adj]." Use operational truth (what they demonstrably do), not aspirations. Every direction must trace back to this sentence.

Step 3. Name analysis

  • Meaning, etymology and translations, including any unwanted meanings in key markets.
  • Sound and rhythm: syllables, hard or soft consonants.
  • Letterforms: distinctive letters, symmetry, ascenders/descenders, double letters, the initial's shape potential.
  • Built-in symbols: objects, actions or places hidden in the name.
  • Length → wordmark viability (short and distinctive favours a wordmark; long favours a symbol or monogram).

Step 4. Brand nouns (30–50)

Mine words from the brand truth and name across these categories (method in references/02-brand-nouns.md): objects · materials · places · actions/verbs · tools · nature · people/roles · cultural references · letters & numbers · abstract qualities · opposites/tensions. Mark each noun ★ if it is ownable (specific to this brand) or ○ if it is generic or category-cliché. Keep clichés visible so you can avoid them.

Step 5. Pairings (10–20)

Strong marks often fuse two ideas into one form. Pair nouns across categories: letter + object, object + action, place + craft, material + tool. For each pairing, give the visual idea in one line ("the S doubles as a rope coil").

Step 6. Directions (4–6)

Pick the strongest pairings and spread them across logo types (wordmark, lettermark, pictorial, abstract, combination, emblem, mascot) using references/03-directions.md. For each direction give:

  • Title (2–4 words)
  • Logo type
  • Concept: what you see
  • Story: one sentence a client can repeat
  • Why it fits: link to the brand truth
  • Risks: cliché, scaling, similarity
  • Techniques: negative space, single line, modular grid, etc.

Step 7. Prompts

For each direction write two prompts using the templates in references/04-prompt-library.md:

  1. Concept sheet: a 3×3 grid of 9 flat black explorations (for divergence).
  2. Single mark: one refined flat mark (for convergence).

Then add 2–3 iteration prompts: "keep X, change Y."

Prompt rules:

  • black on white, flat vector, no gradients, shadows, 3D, mockups or extra text
  • symbols are generated without text; wordmarks are set in real type later
  • never name living designers or existing brands' logos

Adapt the syntax to the user's tool (e.g. Midjourney --ar 1:1 --style raw --no gradient,shadow,text).

Step 8. Avoid list

Category clichés (e.g. coffee: bean, steam, cup; tech: globe, swoosh, circuit, lightbulb), trend gimmicks, and competitor look-alikes.

Output format

Use templates/output.md (human-readable) or templates/output.schema.json (JSON, for tools and apps). Default to Markdown with:

  • headings
  • a noun table
  • one card per direction
  • prompts in fenced code blocks so they're easy to copy

Mode B: Full logo process (when the user wants to go further)

  1. Generate concept sheets with the available image tool. Figma MCP generate_image uses Figma AI credits, so tell the user. Prefer a gpt-image model when letters must render.
  2. Critique with the 9-point rubric and cut hard: references/05-critique-and-stress-tests.md.
  3. Stress-test early versions: 16px, one colour, reversed, blur, embroidery, rotation, look-alike and trademark search (not legal advice).
  4. Refine in vector: construction, optical corrections, reduction threshold, wordmark spacing, small-size cut. See references/06-refine-and-system.md.
  5. Present 2–3 directions in context with the story, recommendation first.
  6. System: lockups, colour, type, pattern, motion idea, usage rules, file exports, credits and ownership notes.

Principles to apply throughout

  1. Truth over trend. If a choice is fashionable rather than true, cut it.
  2. Volume before judgement. Explore many ideas before falling for one.
  3. Two ideas, one form. Fusion beats decoration.
  4. One colour first. If it needs colour to work, it doesn't work.
  5. Test early, test ugly. Small, reversed, blurred, on bad backgrounds.
  6. Reduce to the threshold. Remove until one more cut would lose the meaning.
  7. Design the system, not just the mark. It must extend to icon, pattern, motion and type.
  8. Commit. Recommend one direction clearly. Don't hedge with safe options.

Never

  • Present raw AI output as a final logo.
  • Imitate an existing logo or prompt with brand or designer names.
  • Show more than three options in a client presentation.
  • Skip the brand truth.
  • Claim a mark is legally clear. Always recommend a trademark search and an attorney.