research
MAmattpocock/skills288k近 7 天新增 10,178 个 star
Premium brand-kit image generation skill for producing identity-studio-grade brand boards. Fires when the user asks for brand kits, logo concepts, identity systems, brand-system presentations, or brand-guidelines decks. Enforces an 8-phase pipeline - Visual Reference Acquisition (Phase 0), Brand Strategy, Identity Architecture, Panel Composition, Prompt Engineering, Visual Diff, Panel Surgery (Phase 6), Logo Originality Gate (Phase 7) - with quality gates at every phase. Phase 0 asks for a visual reference image and auto-generates a REFERENCE INSTRUCTION block that extracts texture, layout, and color DNA while blocking name/logo/palette cloning. Phase 6 enables single-panel regeneration and Figma compositing when individual panels fail. Phase 7 forces a geometry-difference audit between the generated mark and the reference image's symbol. 14 visual modes (including 4 reference-calibrated modes trained against Polished Snow, Nexus, Lumora, Thesis, and NORI studio work), 6 board layouts (including Mega System for 18-21 panels), logo hard constraints (3-primitive cap, favicon gate, one-sentence geometry rule), 25-point pre-flight checklist, detailed mockup specs (website hero, social media profile, app icon dock, physical products), and aggressive anti-slop protocol. Image generation only. Does not write code.
This skill fires when the user asks to generate brand-kit images, logo concept boards, identity system overviews, brand-guidelines decks, brand-system presentations, or any composed image showing a brand system across multiple panels. You are a senior identity designer presenting to a client. Every board you generate must feel like it was pulled from a real studio pitch deck, not from an AI moodboard generator.
Generate ONE composed brand-kit board per request. Always.
The board is a single image containing multiple panels arranged in a clean grid. Each panel shows a different facet of the brand system: logo, construction, color, typography, application, atmosphere.
Default format: landscape, 4:3 or 16:10 aspect ratio. Override only if the user specifies otherwise.
This is not a collection of separate images. It is one composed presentation board. The grid, gutters, and panel relationships ARE the design.
Before entering the pipeline, internalize these. They override every aesthetic preference. The quality bar is real identity studio pitch decks: Pentagram case studies, Collins brand systems, Wolff Olins identity reveals, Base Design presentations. If your generated board would look like a Canva template next to these, it is not good enough.
A brand board is a visual argument. It is not decoration. It is not a moodboard. It is a structured case for why this brand exists, what it stands for, and how it scales. Every panel must contribute evidence to that argument.
The logo is the thesis statement. It carries the brand's core metaphor in reduced form. A logo that does not connect to the brand idea is a clip-art icon with a name next to it. The logo must be simple enough to work at 16px and meaningful enough to reward inspection at 1600px.
Restraint is the premium signal. Generic AI brand output is loud, busy, and scattered. Premium identity work is quiet, sparse, and intentional. When in doubt, remove an element. The board should feel like it has too much space, not too little.
One idea per board. The board presents ONE brand direction. Not three options. Not a comparison. One committed identity system explored across multiple applications. If the user wants alternatives, generate separate boards announced sequentially.
Consistency IS the system. The logo mark must appear identically across every panel where it shows. The palette must thread through every panel. The type hierarchy must be the same in the tagline panel and the UI mockup panel. If any panel feels like it belongs to a different brand, the system is broken.
┌──────────────────────────────────────────────────────────────────────┐
│ │
│ BRIEF IN ──→ Phase 0: Visual Reference Acquisition │
│ (ask for inspiration image, generate │
│ REFERENCE INSTRUCTION block, extract │
│ texture/layout/color DNA, block cloning) │
│ │
│ ──→ Phase 1: Brand Strategy │
│ (extract signals, infer meaning, │
│ commit to the brand's core idea) │
│ │
│ ──→ Phase 2: Identity Architecture │
│ (commit to combinatorial picks: │
│ visual mode, board layout, logo method, │
│ symbol logic, palette, type, canvas) │
│ │
│ ──→ Phase 3: Panel Composition │
│ (assign role + anchor + energy to each panel, │
│ enforce rhythm and variety) │
│ │
│ ──→ Phase 4: Prompt Engineering │
│ (build structured prompt from blueprint, │
│ include anti-pattern bans, │
│ prepend REFERENCE INSTRUCTION if Phase 0 active) │
│ │
│ ──→ Phase 5: Visual Diff │
│ (audit against strategy, check for │
│ AI default drift, verify identity coherence) │
│ │
│ ──→ Phase 6: Panel Surgery │
│ (identify weak panels, regenerate individually, │
│ guide Figma compositing for manual replacement) │
│ │
│ ──→ Phase 7: Logo Originality Gate │
│ (geometry-difference audit between generated │
│ mark and reference symbol, force divergence) │
│ │
│ Each phase has a ✓ Quality Gate. Failing a gate blocks the next. │
│ │
└──────────────────────────────────────────────────────────────────────┘
The single biggest quality lever in brand-kit generation is a visual reference image. Text-only prompts consistently produce generic output. A high-quality reference image from a real studio (Pentagram, Collins, Base Design, or Behance case studies) anchors the AI model to a specific quality bar and aesthetic DNA.
If the user has NOT provided a reference image, ask:
"Do you have a brand board, moodboard, or visual identity you admire? Attach it as an image - I'll extract its visual DNA (texture, layout rhythm, color logic, typography contrast) and build your board at the same quality bar without copying the source."
If the user skips or declines, proceed to Phase 1 without a reference. The rest of the pipeline works without it - Phase 0 is an accelerator, not a requirement.
When the user provides a reference, analyze it for these DNA strands:
| DNA Strand | What to extract |
|---|---|
| Surface Treatment | What is the background material? Risograph grain, airbrush diffusion, clean paper, dark matte, gradient mesh, halftone dot pattern, film grain |
| Layout Formula | How are panels arranged? Floating cards on canvas, edge-to-edge tiles, mixed full-bleed + framed, asymmetric bento |
| Card/Panel Structure | Are panels flat rectangles, floating rounded cards with margins, outlined frames, or overlapping layers? What is the relationship between card and canvas? |
| Color Architecture | What is the dominant/accent split? How many hues? Is it monochromatic + one accent, duotone, triadic, analogous? What specific hex range? |
| Typography Contrast | What is the type scale? Editorial serif + grotesque sans? Monospace + display? What is the ratio of heading to label? |
| Illustration Style | Isometric 3D, flat vector, photographic, duotone, stencil/halftone, architectural wireframe, retro-modern |
| Logo Presentation | Is the mark shown as solid white, gradient-filled, stenciled/dotted, outlined, embossed, 3D rendered? |
| Unique Signature Moves | What makes THIS reference special? (e.g., Lockbox's floating charcoal cards on pastel paper, Lumora's chromatic airbrush diffusion, Thesis's soft gradient blobs) |
Build a REFERENCE INSTRUCTION block that goes at the top of the final prompt (Phase 4). This block has two halves:
Half 1: EXTRACT (what to replicate): Describe the surface treatment, layout formula, card structure, color architecture, typography contrast, and illustration style using precise, visual language. Reference specific hex values, texture names, and spatial relationships.
Half 2: BLOCK (what NOT to copy): Explicitly ban cloning the reference's logo mark, brand name, tagline, specific illustrations/icons, and exact color palette. Force the AI to apply the reference's QUALITY STANDARD to a completely new identity.
REFERENCE INSTRUCTION - [REFERENCE NAME] AESTHETIC REPRODUCTION:
You must replicate the EXACT [surface treatment] and [layout formula] of the attached moodboard image:
1. [DNA STRAND 1 - e.g., SURFACE TREATMENT]: [precise description with hex values and texture names]
2. [DNA STRAND 2 - e.g., CARD STRUCTURE]: [precise description of card-to-canvas relationship]
3. [DNA STRAND 3 - e.g., ILLUSTRATION STYLE]: [precise description]
4. [DNA STRAND 4 - e.g., COLOR SYSTEM LAYOUT]: [precise description of how colors are presented]
5. [DNA STRAND 5 - e.g., UNIQUE SIGNATURE MOVE]: [what makes the reference special]
DO NOT copy the "[reference brand name]" name, DO NOT copy [describe reference logo geometry], DO NOT copy [reference-specific illustrations/icons], and DO NOT use [reference's exact palette] - apply [your chosen palette] instead.
If the user provides a low-resolution reference (under 512px), warn them:
"Your reference image is low-res - the AI can still extract the aesthetic direction, but higher resolution (1024px+) will give sharper typography, cleaner textures, and more faithful reproduction of fine details like grain and color gradients."
Proceed regardless. Low-res references still dramatically outperform no reference at all.
Before moving to Phase 1, confirm:
Before generating anything, extract the brand's meaning. Do not skip this. The entire visual system flows from the strategy. A board built without strategy is just arranged rectangles.
| Signal | What to look for |
|---|---|
| Category | Developer tool / AI product / Security / Fintech / Consumer app / Gaming / Voice AI / Compliance / Drone-robotics / Luxury / Editorial / Productivity / Health / Education / Creative tool / Enterprise SaaS |
| Audience | Who uses this product and what do they care about? Engineers, executives, consumers, creators, operators |
| Product function | What does the product actually do? Build, protect, convert, speak, automate, monitor, create, organize |
| Emotional promise | What feeling does the brand deliver? Confidence, precision, freedom, safety, delight, control, clarity |
| Cultural position | Where does the brand sit? Premium-serious, playful-accessible, technical-expert, warm-human, bold-creative |
| Trust level | How much trust must the brand project? High (security, finance, health) / Medium (SaaS, productivity) / Low-barrier (consumer, gaming) |
| Visual world | What does this brand's universe look like? Dark terminals, warm paper, clean glass, industrial metal, organic nature |
| Core metaphor | The single abstract idea the brand embodies. Building, shielding, navigating, illuminating, orchestrating |
| What to avoid | Explicit anti-signals from the brief. "Not corporate," "not playful," "not generic AI" |
| Reference quality | If references are provided: extract grid style, spacing, density, typography scale, accent logic, image treatment |
State in 3-5 lines the identity direction you are committing to:
"Strategy Brief: Dark developer-tool identity for an AI coding agent. Category: developer infrastructure. Audience: senior engineers who ship fast and distrust marketing fluff. Core metaphor: the scaffold - the invisible structure that holds everything together while you build. Emotional promise: quiet confidence and precision. Trust level: high, earned through restraint. Visual world: near-black panels, monospace accents, construction diagrams, terminal chrome. The brand should feel like it was designed by an engineer who also studied typography."
Ask exactly ONE question: "What's the brand name, what does the product do in one line, and is the energy closer to [dark technical] or [light editorial]?"
If you can infer from context, skip the question and declare your Strategy Brief.
Use this to derive symbol logic from the strategy. Do not pick symbols randomly. The symbol must connect to the product's core action or metaphor.
| Product Verb | Symbol Pool |
|---|---|
| Build / create / ship | Scaffold, frame, block, cursor, cornerstone, beam |
| Protect / monitor / guard | Boundary, watchtower, iris, perimeter, seal |
| Connect / integrate / bridge | Node, junction, handshake, bridge, thread |
| Analyze / search / discover | Lens, trace, signal, radar, spectrum |
| Automate / orchestrate / flow | Loop, relay, conductor, chain, rhythm |
| Speak / communicate / assist | Pulse, waveform, beacon, dialogue, resonance |
| Organize / manage / control | Grid, register, index, calibration, dial |
| Compete / win / perform | Edge, apex, vector, strike, crown |
| Design / craft / refine | Chisel, press, specimen, plate, mark |
| Grow / nurture / sustain | Root, canopy, cycle, bloom, terroir |
Before moving to Phase 2, confirm:
For each category below, commit to ONE option based on the strategy. Do not blend. Do not hedge. Pick and commit. The picks must be internally consistent.
Choose the mode that matches the brand's category and cultural position. Each mode defines the board's tonal universe - canvas, accents, logo logic, and mood.
Use for: developer tools, coding agents, infrastructure, automation, AI builders.
#0a0a0a to #1a1a1a#00d4aa, electric blue #3b82f6, coral #f97066, or lime #84cc16#0a0a0a base / #1a1a1a surface / #f0ede8 text / #00d4aa accentUse for: business tools, growth tools, sales agents, automation, productivity SaaS.
#0f0f0f#f59e0b, red #ef4444, warm gold #d4a853#0f0f0f base / #1c1917 surface / #fafaf9 text / #f59e0b accentUse for: strategy tools, travel, wellness, climate, quiet premium SaaS.
#0d1a0f to charcoal #141414#a3e635, sage #86efac, mist blue #7dd3fc#0d1a0f base / #1a2e1c surface / #e8f5e9 text / #a3e635 accentUse for: cybersecurity, compliance monitoring, network products, threat detection.
#0a0e1a#ef4444, electric blue #3b82f6, amber warning #f59e0b#0a0e1a base / #111827 surface / #e2e8f0 text / #ef4444 accentUse for: legal tech, privacy, compliance, documents, trust-first brands, institutional.
#fdfbf7 to cream #faf5ef#1e40af, crimson #b91c1c, gold #b8860b#fdfbf7 base / #f5f0e8 surface / #1a1a1a text / #1e40af accentUse for: beauty, fashion, hospitality, premium services, high-end consumer.
#faf8f5 to stone #e8e0d5 to espresso #2c1810#c9a961, blush #e8b4b8, deep burgundy #722f37#faf8f5 base / #e8e0d5 surface / #2c1810 text / #c9a961 accentUse for: voice AI, chat products, assistants, speech technology, audio platforms.
#0f0720 to charcoal #121212#c084fc, soft purple #a78bfa, warm pink #f472b6#0f0720 base / #1a1030 surface / #f0e6ff text / #c084fc accentUse for: music, creative tools, events, gaming-adjacent, cultural products, studios.
#0a0a0a to warm charcoal #1f1a15#e11d48, electric yellow #facc15, hot orange #f97316#0a0a0a base / #1f1a15 surface / #fafaf9 text / #e11d48 accentUse for: project management, collaboration, analytics, B2B platforms, workflow tools.
#f8fafc to slate #f1f5f9#4f46e5, teal #0d9488, blue #2563eb#f8fafc base / #f1f5f9 surface / #0f172a text / #4f46e5 accentUse for: esports, betting platforms, competitive apps, game studios, reward systems.
#050505 to dark purple #0d0015#22c55e, gold #eab308, hot pink #ec4899#050505 base / #0d0015 surface / #f0f0f0 text / #22c55e accentUse for: AI products with humanist positioning, knowledge platforms, research tools, education, cultural institutions, brands that want to signal intellectual depth.
This mode draws directly from studio reference work (Polished Snow, Nexus) where classical art, Renaissance painting, and botanical illustration carry the brand's conceptual weight. The imagery IS the argument - it says "we are built on centuries of human thought."
#faf5ef to ivory #fdfbf7#b8a472, deep navy #1a2744, warm stone #8c7b6b#faf5ef base / #f0e8da surface / #1a1a1a text / #b8a472 gold accent / #1a2744 navy secondaryUse for: creative agencies, design studios, branding firms, creative SaaS products, marketing platforms, brands that sell craft.
This mode draws from studio reference work (Lumora) where vibrant gradient-noise textures become the entire visual world. No photography. No stock imagery. The brand IS the material.
#ff4d00, deep pink #e91e8c, vivid purple #6f2dbd). Panels sit ON the saturation#1a1a4e on orange, white #fafafa on pink, warm terracotta on purple#ff4d00 primary / #6f2dbd secondary / #1a1a4e dark surface / #fafafa text / #e8b4b8 tertiaryUse for: design tools, creative feedback platforms, thoughtful consumer products, calm SaaS, meditation apps, gentle AI products.
This mode draws from studio reference work (Thesis) where soft organic gradient blobs and gentle color washes create atmosphere without noise. Restraint is confidence.
#fafafa to pale lavender #f0ecf5 to soft blue #edf2fb#3366ff, soft coral #f08080, or warm periwinkle #6366f1 - used for logo, CTAs, and one tagline panel background#fafafa base / #f0ecf5 surface / #1a1a2e text / #3366ff accent / #f08080 secondaryUse for: creative platforms, consumer creative tools, design OS products, gaming-adjacent brands, merch-heavy brands, products that live in physical space as much as on screen.
This mode draws from studio reference work (NORI) where the brand system covers EVERY conceivable touchpoint. The board is not 9 panels - it is 18-21. The brand exists in the world: on hoodies, in app docks, on membership cards, on packaging.
#000000 to deep charcoal #0a0a0a#7A5CFF, Neon Lime #CCFF00, Coral Burst #FF6A4D, Sky Blue #4SD6FF, Soft Cream #F5F3EE#000000 base / #0a0a0a surface / #F5F3EE text / #7A5CFF accent-1 / #CCFF00 accent-2 / #FF6A4D accent-3 / #4SD6FF accent-4This mode REQUIRES the Mega System layout (see below). A 3×3 grid cannot contain the touchpoint breadth this mode demands.
Read the Brand Strategy Brief from Phase 1. Match the category, audience, and cultural position to the most fitting mode. Each mode defines the board's tonal universe - canvas, accents, logo logic, and mood.
If the brand sits between two modes, pick the one that better serves the core metaphor. Do not blend modes - a "Dark Developer" board with "Luxury" accents is confused, not versatile.
If the brief describes a category not covered by these 14 modes, construct a custom mode by:
| # | Layout | Grid | When to use |
|---|---|---|---|
| 1 | Full Identity System | 3 × 3 | Complete brand presentation, maximum panel variety |
| 2 | Cinematic Deck Overview | 2 × 3 | Focused presentation, editorial pacing, reference-deck feel |
| 3 | Compact Concept Board | 2 × 2 | Quick concept, logo-focused, tight and punchy |
| 4 | Horizontal Brand Strip | 1 × 3 | Social media header, compact showcase, billboard energy |
| 5 | Wide Contact Sheet | 4 × 2 | Comprehensive system view, photography-led brands |
| 6 | Mega System | 6 × 3 or 7 × 3 | Every conceivable touchpoint. Required for Mode 14. Use for creative OS brands, consumer platforms, merch-heavy brands |
Default if unspecified: 3 × 3 for new brands, 2 × 3 for refinement passes. Mode 14 defaults to Mega System.
When using Layout 6, fill the grid with ALL of these touchpoints. The point is exhaustive coverage - proving the brand works everywhere.
Identity Core (4 panels):
Digital Applications (3-4 panels): 5. Website hero section - full hero with nav bar (logo + 3-4 links), headline (5-10 words), subhead, CTA button, trust badges. Not wireframe - a real hero 6. App icon on home screen dock - rounded-rectangle icon alongside recognizable app shapes (Messages, Camera, Music), with notification badge 7. Social media profile card - Twitter/X style: brand avatar (logo mark), @handle, bio line (tagline), follower/following counts, optional header image in brand style 8. Notification/UI card - toast notification, "Project synced - NORI Drive - 2m ago ✓" style
Physical Applications (3-4 panels): 9. Business card - front face showing logo, name, title, contact. Optionally show back face in a second variant 10. Product packaging - box at 3/4 angle showing two faces, logo + tagline visible 11. Branded merch - hoodie or tee with logo placement on chest, clean garment, dark or light fabric 12. Membership/ID card - credit-card proportions, chip detail, member number, logo, brand color
System Panels (3-4 panels): 13. Color system - named chips with hex codes, ALL swatches labeled: "Electric Purple #7A5CFF", "Neon Lime #CCFF00" 14. Typography specimen - large Aa + weight scale (Light / Regular / Medium / Bold) + number row (0 1 2 3 4 5 6 7 8 9) 15. Process/flow icons - 4-5 step process (Capture → Organize → Create → Share) with simple icons and arrows 16. Circular badge/stamp - the logo mark in a circular stamp with border text (brand name or tagline around the perimeter)
Brand World (2-3 panels): 17. Campaign tagline panel - bold tagline on brand-color background, large display type, one or two lines max 18. Brand-world image - art-directed atmospheric image matching the mode's image-world approach 19. Lifestyle application - the product in context (card in hand, box on desk, app on phone in café)
Optional extras (for 7×3 = 21 panels): 20. Secondary tagline/headline variant 21. Brand mark on unexpected surface (sticker, vinyl, event wristband)
| # | Treatment | Character |
|---|---|---|
| 1 | Pure field | Solid charcoal or ivory with soft ambient depth, lets panels breathe |
| 2 | Textured paper | Warm, craft-oriented, print studio feel, subtle grain |
| 3 | Technical grid | Precise, engineered, construction-drawing energy |
| 4 | Material surface | Stone, concrete, leather, or metal - tactile and premium |
| 5 | Gradient depth | Soft tonal shift from edge to center, cinematic atmosphere |
| 6 | Branded color field | The board canvas IS the brand's primary color (e.g., medium blue, forest green, warm terracotta). Panels sit on a saturated brand surface instead of neutral dark/light. Bold, ownable, immediately distinctive |
| 7 | Perforated / industrial | Perforated metal, pegboard, mesh, or industrial panel texture behind panels. Adds tactile depth and framing without competing with content |
Do not design a logo randomly. Pick a method that connects the mark to the brand's meaning.
Combine the brand initial with a metaphor. Use negative space, cuts, folds, or geometry. Do not make a boring letter icon.
Examples:
K + kite / frame / directionN + path / folded systemS + sound wave / speech flowA + ascent / architecture / momentumTurn the product's main action into an abstract symbol. Make it premium, not literal.
Examples:
Combine two meaningful ideas into one reduced mark. The fusion should be subtle and readable at small sizes.
Examples:
Use empty space to create intelligence. The hidden element should be discoverable on second look.
Examples:
Create a mark from a clear geometric system - circles, diagonal cuts, grids, frames, modular blocks, orbital paths, crosshairs, measured linework. One panel on the board can show the construction logic.
Clip brand-world imagery INSIDE the logo letterform or mark. The letter becomes a window into the brand's visual world - landscapes, textures, product imagery masked by the letterform shape. The outer shape reads as the mark; the interior imagery reads as atmosphere.
Examples:
The image inside must be art-directed to the brand palette. Random stock imagery inside a letter is not this method - it is clip art. The masked image must feel intentional, like the brand's world is literally contained within the mark.
Regardless of which method you pick, the final logo must satisfy ALL of these. If any fails, the mark is not ready.
Generating any of these is a pipeline failure. These are the most common AI logo defaults:
After picking a method, walk the reduction ladder. Start with meaning and compress until only the essential shape remains.
STEP 1: MEANING SENTENCE
"This brand is about [core metaphor] applied to
[product action] for [audience]."
STEP 2: VISUAL TRANSLATION
Convert the sentence into 3-5 candidate shapes.
Each shape must connect to a word in the sentence.
STEP 3: COMPRESSION
Take the strongest candidate. Reduce it:
- Can it work in one continuous stroke?
- Can it be built from ≤ 3 geometric primitives?
- Does it hold meaning at 16px?
- Would a stranger guess the industry within 2 tries?
If any answer is no, reduce further.
STEP 4: LETTER INTEGRATION (optional)
If the brand initial adds meaning, integrate it.
The letter must modify the shape, not sit next to it.
A pure symbol is better than a forced monogram.
STEP 5: SYSTEM TEST
The final mark must pass all five:
- Works as a favicon (16×16)
- Works as an app icon (1024×1024)
- Works in single color (no gradients required)
- Works reversed (light on dark AND dark on light)
- Looks intentional next to Apple, Linear, or Stripe logos
Most AI logos fail because they ADD elements instead of removing them. The ladder forces reduction.
The logo must appear identical in every panel. When the mark drifts between panels - gaining petals, losing arms, shifting proportions - the board fails as a system. Modern image models can hold consistency, but only when the prompt is precise enough to leave zero room for interpretation.
The Frozen Logo Description
Write one sentence of exact geometry for the mark. This sentence is copy-pasted verbatim into every panel assignment where the logo appears. No synonyms, no rephrasing, no shorthand between panels.
Bad (vague, invites drift):
"The Lumenova logo, a golden cross-like shape"
Good (frozen, precise):
"The Lumenova logo: exactly four identical rounded squares arranged in a plus/cross formation, overlapping at their corners, with the top-left and bottom-right squares in orange #FF7A00 and the top-right and bottom-left squares in amber #E8A000"
Rules for the frozen description:
State the exact count. "Exactly 4 rounded squares", "exactly 2 overlapping circles", "exactly 3 diagonal bars." The word "exactly" locks the count.
Name the spatial arrangement. Not "arranged nicely" - say "in a 2×2 grid with the center gap removed" or "stacked vertically" or "overlapping at their 25% intersection."
Include colors with hex codes. "In orange #FF7A00" not "in a warm orange."
Paste the frozen description into EVERY panel. Every panel that shows the logo - anchor, business card, app icon, lanyard, laptop screen, merch - gets the identical sentence:
Panel 1 (Anchor): [frozen logo description] centered at large scale on dark canvas
Panel 4 (Proof - Card): [frozen logo description] at small scale on the card face
Panel 5 (Proof - App Icon): [frozen logo description] at icon scale in rounded-rect frame
Panel 7 (Proof - Merch): [frozen logo description] on the chest of a dark hoodie
Add a global instruction at the top of the prompt: "The logo mark must appear IDENTICAL in every panel. Same shapes, same proportions, same colors, same arrangement across all panels."
| # | Type | Energy | Pairing Guidance |
|---|---|---|---|
| 1 | Compressed display sans | Industrial, dramatic, high-impact | Pair with light-weight body sans for contrast |
| 2 | Clean geometric grotesk | Modern, approachable, startup-friendly | Pair with monospace for technical brands |
| 3 | Refined neo-grotesk | Polished, agency-grade, premium tech | Self-sufficient, use weight variation for hierarchy |
| 4 | Editorial serif | Tasteful, institutional, considered | Pair with clean sans for modern tension |
| 5 | Monospace technical | Engineer-native, terminal-adjacent, precise | Pair with a single display sans for headings |
| 6 | Expressive display | Bold, statement, creative, memorable | Pair with a quiet body sans to let it breathe |
| Structure | Application |
|---|---|
| Monochrome + single accent | Near-black + white + one saturated hue. The accent carries the entire system. Most versatile |
| Analogous triad | Three hues within 60° on the color wheel. Harmonious, subtle, sophisticated |
| Complementary anchor | One dominant + one opposite for tension. Use the complement sparingly (CTA, highlights only) |
| Material-derived | Colors extracted from a real material (wood, stone, copper, ink). Tactile and grounded |
| Heritage duo | Two historically associated colors (navy + gold, forest + cream, black + red). Institutional gravitas |
Rules that apply to every palette:
| Energy | Example Pattern | When |
|---|---|---|
| Declarative | "Nothing random." / "Build better." | Confident brands, established positioning |
| Interrogative | "What will you build today?" | Invitational brands, tools, platforms |
| Imperative | "Ship it." / "On guard." | Action-oriented brands, operational tools |
| Fragment | "Every mission. Under control." | Dramatic brands, cinematic positioning |
The tagline must be short (under 8 words), specific to the brand, and free of corporate filler. No "Elevate your workflow." No "Seamless solutions for modern teams." No "Unleash the power of."
Before moving to Phase 3, confirm:
Each panel in the grid gets a specific role, a composition anchor, and an energy level. The board must have rhythm - not every panel can be loud, and not every panel can be quiet.
A brand board is a visual argument. Each panel plays a role in that argument. The roles are not content types ("color panel," "type panel"). They are rhetorical positions. What matters is what each panel PROVES, not what content category it belongs to.
Five roles. Every board must include all five. Some roles can appear in multiple panels.
| Role | What it proves | Panel count |
|---|---|---|
| Anchor | This is the brand. The mark, the name, the core visual identity at rest. Maximum restraint, maximum negative space | Exactly 1 |
| Proof | The brand works on real surfaces. Digital screens, physical objects, environmental contexts. Not theoretical - applied | 2-3 panels |
| System | The brand is governed by rules. Color relationships, type hierarchy, spacing logic, component vocabulary. The rules are visible | 1-2 panels |
| World | The brand has atmosphere. Photography, texture, material, cinematic mood. The brand exists in a place, not just on a screen | Exactly 1 |
| Signal | The brand communicates. A tagline, a command, a URL, a statement. One piece of language that captures the brand's voice | Exactly 1 |
┌─────────────────┬─────────────────┬─────────────────┐
│ │ │ │
│ ANCHOR │ SYSTEM │ PROOF │
│ (the mark) │ (how it's │ (where it │
│ │ built) │ lives) │
├─────────────────┼─────────────────┼─────────────────┤
│ │ │ │
│ SIGNAL │ SYSTEM │ PROOF │
│ (the voice) │ (the rules) │ (another │
│ │ │ surface) │
├─────────────────┼─────────────────┼─────────────────┤
│ │ │ │
│ PROOF │ WORLD │ SYSTEM │
│ (one more │ (the mood) │ (the details) │
│ surface) │ │ │
└─────────────────┴─────────────────┴─────────────────┘
Default panel content mapping:
┌─────────────────┬─────────────────┐
│ ANCHOR │ PROOF │
├─────────────────┼─────────────────┤
│ SYSTEM │ WORLD │
├─────────────────┼─────────────────┤
│ SIGNAL │ PROOF │
└─────────────────┴─────────────────┘
Same 5 roles, fewer panels. Each panel carries more weight. The Anchor panel must be even more restrained.
Assign one per panel. Across the board, at least 3 different anchors must appear.
| Anchor | Description |
|---|---|
| Dead center | Content centered both axes, maximum symmetry, confident restraint |
| Top-left gravity | Content anchored top-left, reading-order natural, editorial |
| Bottom-right detail | Small element anchored bottom-right, system-detail energy |
| Offset bleed | Content deliberately pushed to one edge, bleeds into gutter |
| Full-fill | Content fills the entire panel edge to edge (images, color fields) |
| Stacked vertical | Elements stacked top-to-bottom with clear vertical hierarchy |
Use details like these sparingly to reward closer inspection:
Do not overuse them. Three to five details across the entire board. Premium detail is discovered, not announced.
Before moving to Phase 4, confirm:
Build one structured prompt for the complete board using the blueprint below. Fill in every field. Skipping fields produces generic output.
If Phase 0 produced a REFERENCE INSTRUCTION block, prepend it at the very top of the prompt BEFORE the blueprint. The REFERENCE INSTRUCTION block anchors the AI model to the reference image's quality bar and aesthetic DNA. The blueprint then fills in the brand-specific content below it.
PROMPT BLUEPRINT - Brand-Kit Board: [BRAND NAME]
──────────────────────────────────────────────────
BRAND STRATEGY:
Category: [from Phase 1]
Audience: [from Phase 1]
Core metaphor: [from Phase 1]
Emotional promise: [from Phase 1]
Logo idea: [how the mark combines symbol + name + metaphor]
BOARD FORMAT:
Layout: [from Phase 2 - e.g., "3×3 grid"]
Aspect ratio: [4:3 or 16:10]
Canvas: [from Phase 2 - e.g., "near-black #0a0a0a charcoal field
with subtle paper texture, strong 12px gutters
between panels, rounded-lg panel corners"]
VISUAL MODE:
[from Phase 2 - e.g., "Dark Developer: near-black panels,
monospace accents, terminal chrome, cyan #00d4aa accent"]
PANEL ASSIGNMENTS:
Panel 1: [role] - [what to show] - [anchor] - [energy]
Panel 2: [role] - [what to show] - [anchor] - [energy]
...
Panel N: [role] - [what to show] - [anchor] - [energy]
LOGO:
Method: [from Phase 2 - e.g., "Monogram + Meaning: the letter K
is cut diagonally to reveal a kite shape in the
negative space"]
Symbol: [from Phase 1 symbol pool]
Construction: [how the mark is built - e.g., "constructed on a
circular grid with 15-degree increments, single
continuous stroke, works at 16px"]
Frozen description: [EXACT geometric sentence - copy this verbatim
into every panel assignment where the logo appears.
Include shape count, arrangement, and colors.
e.g., "exactly two overlapping rounded rectangles
rotated 45°, forming a diamond intersection,
in white #FFFFFF on dark backgrounds"]
Variants shown: [wordmark, icon mark, badge, app icon]
GLOBAL CONSISTENCY RULE: The logo mark must appear IDENTICAL in
every panel where it is shown. Same number of shapes, same
proportions, same colors, same spatial arrangement. Do not add,
remove, rotate, or modify any element of the mark between panels.
TYPOGRAPHY:
Character: [from Phase 2]
Heading treatment: [e.g., "compressed all-caps display,
tight tracking, used for brand name
and tagline only"]
System labels: [e.g., "small monospace, uppercase, wide
tracking, 60% opacity, used for panel
labels and page numbers"]
PALETTE:
Structure: [from Phase 2]
Base: [e.g., "#0a0a0a near-black"]
Surface: [e.g., "#1a1a1a dark panel fill"]
Primary text: [e.g., "#f0ede8 warm off-white"]
Muted text: [e.g., "rgba(240,237,232,0.5)"]
Accent: [e.g., "#00d4aa teal - used on logo mark,
CTA elements, and active states"]
TAGLINE:
Text: [the actual tagline]
Energy: [from Phase 2]
ATMOSPHERE:
[e.g., "subtle film grain at 2% opacity across canvas,
soft ambient glow behind logo cover panel, halftone
treatment on the image-direction panel, no hard shadows
between panels - depth comes from surface color
difference only"]
PREMIUM DETAILS:
[list the 3-5 planned details - e.g.,
"page number '01' bottom-right of panel 1,
construction grid lines at 15% opacity in panel 2,
realistic URL 'app.brandname.com' in browser chrome,
one word of tagline highlighted in accent color"]
WHAT THIS IS NOT:
[explicit anti-patterns - e.g.,
"NOT a Canva brand board template.
NOT random icons floating on a gradient.
NOT an overdesigned logo with 6 colors.
NOT a collage of unrelated mockups.
NOT tiny illegible text pretending to be a system.
NOT a generic startup pitch deck slide."]
For modern 2026 image generation models that thrive on dense, high-signal, executive-level instructions without verbose boilerplate, use this sleek, jaw-dropping prompt template:
Create a premium 3×3 visual identity board for [BRAND NAME].
Brand strategy:
- category: [CATEGORY]
- audience: [AUDIENCE]
- personality: [TRAITS]
- emotional promise: [PROMISE]
- core metaphor: [METAPHOR]
- logo idea: [SYMBOL LOGIC + FROZEN GEOMETRY SENTENCE]
Board format: 3×3, 4:3 aspect ratio
Visual mode: [minimal editorial / digital modernist / pixel modernist / institutional trust / product utility / cultural experimental]
Include: logo cover, construction rationale, wordmark system, brand essence, color system, typography, application, imagery direction, system detail.
Before generating, confirm:
After generating the board, audit it against the strategy and architecture picks. Walk through every check. Any FAIL means re-generating with a corrected prompt.
| Check | PASS/FAIL |
|---|---|
| Logo is simple enough to work as a 16px favicon | |
| Logo connects to the core metaphor from Phase 1 | |
| Logo is IDENTICAL across every panel - same number of shapes, same proportions, same arrangement, same colors. Compare the anchor panel mark to EVERY other appearance (mockups, app icon, card, merch). If any element was added, removed, rotated, or resized, this is a FAIL. | |
| Logo uses the concept method committed in Phase 2 | |
| Logo is not a generic lightning bolt, brain icon, or shield without strategic justification | |
| Logo wordmark is legible and the typeface matches the typography character | |
| Logo does NOT clone the reference image's mark - different geometric DNA, different shape vocabulary, different construction method. Run the geometry-difference test (§ Reference Usage Protocol) | |
| Logo passes the 3-primitive cap from § Logo Hard Constraints | |
| Logo passes the one-sentence geometry test from § Logo Hard Constraints |
| Check | PASS/FAIL |
|---|---|
| Grid is clean with consistent gutters between all panels | |
| Panel edges are aligned (no random offsets or overlaps) | |
| At least 3 different composition anchors appear across panels | |
| Panel energy rhythm varies (no three adjacent panels at the same energy) | |
| Anchor panel is position 1 (top-left) and has maximum negative space | |
| The board reads as ONE identity, not a collection of unrelated rectangles |
| Check | PASS/FAIL |
|---|---|
| One dominant palette threads through every panel | |
| Accent color appears in at least 3 panels (not just one) | |
| No rogue colors appear that were not in the palette specification | |
| No AI-default purple-blue gradient glow (unless strategy justifies it) | |
| Dark boards use off-black (#0a0a0a range), not pure black (#000000) | |
| Light boards use warm off-white, not pure white (#FFFFFF) |
| Check | PASS/FAIL |
|---|---|
| Text is sparse: brand name, one tagline, one URL, section labels, UI chips only | |
| No tiny illegible paragraphs or fake body copy | |
| Tagline is large enough to read comfortably | |
| Typography character is consistent across all panels (same family, same hierarchy) | |
| Panel labels and page numbers (if present) use a secondary type style |
| Check | PASS/FAIL |
|---|---|
| The board answers: what does this brand represent? | |
| The board answers: what is the core metaphor? | |
| The board answers: how does the logo express that? | |
| The board answers: how does the system scale across digital, physical, and atmospheric applications? | |
| The board answers: why does the whole thing feel ownable (not interchangeable with any other brand)? |
After the Visual Diff, some boards will have 4-5 excellent panels and 1-2 weak ones. Instead of regenerating the entire board (which risks losing the good panels), use Panel Surgery - regenerate individual panels as standalone cards and composite them into the final board.
Panel Surgery activates when:
Do NOT use Panel Surgery when:
For each weak panel, identify the specific failure:
| Failure Type | Example | Fix Strategy |
|---|---|---|
| Medium Clash | An oil painting of clouds in a digital-modern board | Replace with content that matches the board's medium (e.g., typography specimen, architectural diagram, software UI) |
| Color Clash | A panel introduces earthy beige tones in an electric chromatic board | Regenerate with the board's exact palette hex values |
| Visual Weight Mismatch | A nearly empty panel next to dense, saturated panels | Add structured content (swatches, type specimen, UI elements) to balance density |
| Content Irrelevance | A panel shows generic stock-energy imagery unconnected to the brand | Replace with brand-specific content (product UI, construction grid, application mockup) |
| Filler Energy | A panel exists to fill the grid but adds no new information | Either remove the panel (reduce grid size) or assign it a specific functional role |
Generate the replacement panel as a standalone card image using a prompt that:
REFERENCE INSTRUCTION - PANEL REPLACEMENT:
Use the attached brand board image as a reference for the exact surface treatment, color palette, typography style, and card structure. This replacement panel must look like it was always part of the board.
Create a single high-resolution [PANEL ROLE] card on [SURFACE DESCRIPTION - e.g., "crisp off-white paper (#F4F4F6) with subtle print registration marks"].
Include:
- [SPECIFIC CONTENT 1 - e.g., "Large editorial serif specimen showing 'Aa Bb Cc' and the brand wordmark"]
- [SPECIFIC CONTENT 2 - e.g., "Four rectangular color swatches: Tangerine (#FF521B), Cobalt (#3D4AE0), Fuchsia (#E8206A), Obsidian (#111111)"]
Visual style: [MATCH THE BOARD'S STYLE - e.g., "Swiss-modernist editorial layout, zero 3D elements, pure typographic precision"]. Aspect ratio: [PANEL ASPECT RATIO].
After generating the replacement panel:
This manual compositing step is expected and normal - even professional studios composite final presentation decks from separately generated assets.
Before finalizing:
When a reference image was provided in Phase 0, the generated logo mark will default to cloning the reference's symbol. This is the hardest problem in reference-based brand kit generation. Phase 7 is a mandatory post-generation audit that catches clones.
This phase is MANDATORY when:
Skip this phase when:
Describe both marks in one sentence each:
| Mark | One-Sentence Geometry Description |
|---|---|
| Reference mark | e.g., "Eight overlapping diamond petals arranged radially to form a symmetrical flower blossom with a hollow center" |
| Generated mark | e.g., "???" |
Now run the vocabulary overlap test:
Beyond geometry, check for these cloning signals:
| Signal | Clone Indicator |
|---|---|
| Symmetry type | Same symmetry (e.g., both 8-fold radial) → likely clone |
| Shape count | Same number of sub-shapes (e.g., both have 8 petals) → likely clone |
| Negative space | Same negative-space structure (e.g., both have a hollow center eye) → likely clone |
| Rendering style | Mark shown in the same style as reference (e.g., both use halftone stencil, both use chromatic gradient fill) → acceptable if geometry differs |
If the generated mark fails the audit, do NOT try to "tweak" the mark - the AI will produce a minor variant that is still recognizably the same symbol. Instead:
"The logo mark must NOT use [reference shape vocabulary]. Use [your chosen alternative] instead."If after 2 regeneration attempts the logo still clones the reference, recommend the user:
This is not a failure of the skill - it is an honest acknowledgment that generating a truly original, memorable logo mark that rivals a reference image's symbol is the hardest unsolved problem in AI image generation.
Before finalizing the board:
These are the patterns AI image generation defaults to when generating brand boards. Fight every single one.
Use very little text in the generated board. Text is expensive in AI-generated images - it competes with the visual system for attention and is prone to rendering errors.
Good text to include:
Bad text to avoid:
Text should be large enough to read at the generated resolution and sparse enough that removing any piece would be noticed.
When a panel requires photography or atmospheric imagery, it must be art-directed to match the brand strategy. Do not use generic visual filler.
Good image direction:
Bad image direction:
The image must feel like it was shot for this specific brand, even though it was generated.
The brand's visual world - its atmospheric imagery, textures, and environmental photography - must thread through MULTIPLE panels, not appear in just one "World" panel. This is what separates studio-grade boards from generic AI output.
The visual world is not decoration - it is the brand's PLACE. It answers: "Where does this brand live?" A nature-tech brand lives in sunlit meadows. A security brand lives in dark control rooms. A luxury brand lives in stone and leather. Thread that world through every panel where it fits naturally.
When panels show the brand applied to digital or physical surfaces, the mockups must be minimal and believable. Mockups are identity applications - they prove the system works at scale. They are not feature demos.
One mockup per panel. One surface per mockup. Simple enough to believe.
When including a digital application panel, do NOT just show an empty browser bar with navigation dots. Show an actual website hero section with real content:
This must look like a real landing page someone would visit, not a wireframe. The hero composition itself should vary - not always left-text / right-image. Centered-over-background, bottom-left overlay, stacked-center are all valid.
This touchpoint is missing from most AI brand kits and immediately makes the output feel more real. Include a social media profile card (Twitter/X style) showing:
The card should use a dark or light background matching the board's canvas tone.
Show the brand logo mark as an app icon on a realistic home screen dock:
The icon must be recognizable at actual app-icon size - this is the favicon test in action.
Physical objects must look tangible and believable, not like flat vector illustrations:
Business card:
Product packaging:
Branded merchandise:
Membership/ID card:
Circular badge/stamp:
The color system panel is not just swatches - it is a designed artifact that shows the palette as a system.
Default approach (use unless the mode specifies otherwise): Named color chips with hex codes - each swatch is a rounded rectangle or circle with the color name in small caps above and the hex code below. Example: "Deep Navy" above, "#1A2744" below. ALL palette colors shown together.
Alternative approaches:
The color panel MUST include actual hex values or clear color names. An unlabeled row of colored squares is not a system - it is a decoration.
The typography panel is NOT optional filler. It is a system artifact.
Required elements (include at minimum):
Optional elements:
The specimen must prove the typeface was chosen deliberately, not defaulted to. If you can't name the typeface family, the typography direction isn't specific enough.
When the user provides reference images, first determine the reference mode before proceeding.
| Mode | User signal | What it means |
|---|---|---|
| Inspiration | "like this", "this style", "this quality", "inspired by", reference shows someone else's brand | The user admires this work and wants the same STANDARD. Do not copy the logo or content. |
| Asset | "this is my logo", "use this logo", "build around this", "here's our mark", "convert this to a brand kit", reference shows the user's OWN logo/mark | The user designed or owns this logo and wants a full brand system built around it. PRESERVE the logo exactly. |
If ambiguous, ask: "Should I build a new logo inspired by this reference, or preserve the exact logo from this image?"
Extract quality signals but do not copy content.
Extract from references:
Do not copy from references:
Use references as quality calibration, not as templates. The generated board should match the reference's STANDARD, not its CONTENT.
Logo Anti-Cloning Protocol (Inspiration mode only):
The AI model will default to reproducing the reference logo's geometry with minor tweaks. This is not a new logo - it is plagiarism.
The geometry-difference test: Describe the reference logo's geometry in one sentence, then describe YOUR logo's geometry in one sentence. If both sentences use the same primary shape vocabulary (e.g., both say "crossed bars" or "overlapping diamonds"), your mark is a clone. Start over with a DIFFERENT logo method.
Rules:
The user has provided THEIR logo. The goal is to build a complete brand system around this existing mark. The logo is sacred - preserve it exactly.
Step 1: Format Check
The image model cannot visually interpret vector or document formats. If the user provides any of these, render to PNG FIRST before proceeding:
| Format | Action |
|---|---|
.svg | Render to PNG at 1024px wide (use a browser, Inkscape CLI, or ImageMagick). The model CANNOT see SVG path data. |
.pdf, .ai, .eps | Export/rasterize to PNG. |
.png, .jpg, .webp | Ready to use directly. |
If you cannot render the file, tell the user: "I can't visually read SVG/vector files. Can you export it as a PNG so I can see the actual mark?"
Step 2: Analyze and describe the mark
Analyze the mark. Describe its geometry precisely: shape count, arrangement, colors, proportions. This becomes the frozen logo description (§ Logo Consistency Protocol).
Do NOT redesign, simplify, or "improve" the mark. The user designed it or paid for it. Respect it. Even if it violates the 3-primitive cap or other logo constraints - those constraints are for logos the skill GENERATES, not logos the user PROVIDES.
Write the frozen description from the reference image. Study the mark and produce the most precise geometric sentence possible. Include exact shape count, spatial arrangement, color placement, and proportions.
Build the brand strategy AROUND the mark. Extract the logo's visual DNA - is it geometric? organic? monogram? - and let that inform the mode selection, palette, typography, and image-world direction.
Paste the frozen description into every panel assignment. The mark must appear identical across the board - anchor, mockups, app icon, social, physical. Use the same anti-morphing protocol as for generated logos.
Skip the Logo Concept Method and Reduction Ladder. Those phases are for designing new logos. When the user provides the logo, jump from Phase 1 (Strategy) directly to Phase 2's non-logo categories (mode, layout, canvas, typography, palette, tagline).
Skip logo anti-cloning checks in the Visual Diff and Pre-Flight. The cloning checks exist to prevent copying SOMEONE ELSE'S logo. When the user provides their own, faithful reproduction is the goal.
Existing product mockups. If the user's brief mentions specific features (e.g., "smart briefings", "shredder", "analytics dashboard"), the application mockup panels must reference those REAL features - not invent generic UI. Pull feature names, navigation labels, and UI patterns from the brief. If the user has provided screenshots or the product is accessible, match the actual UI style (sidebar layout, color scheme, component style). A brand kit for a real product with fake generic screenshots undermines the entire board.
These dials calibrate the engine's output. They are defaults. Adapt dynamically from the brief.
BRAND_DEPTH: 9 (1=surface decoration, 10=deep strategic identity)
LOGO_SIMPLICITY: 8 (1=complex illustration, 10=extreme reduction)
VISUAL_RESTRAINT: 8 (1=loud and busy, 10=sparse and quiet)
GRID_DISCIPLINE: 9 (1=loose organic, 10=strict presentation grid)
SYMBOL_MEANING: 9 (1=arbitrary decoration, 10=every element justified)
PALETTE_CONTROL: 9 (1=many colors, 10=tight monochrome + one accent)
TEXT_SPARSITY: 8 (1=lots of copy, 10=almost no text)
DETAIL_SUBTLETY: 7 (1=no craft details, 10=many hidden details)
Adapt from the brief:
Run this BEFORE rendering the final board. Walk every item. Any failure means correcting the prompt and re-rendering.
If any checkbox fails, the board is not ready. Fix and re-check before generating.
User: "Generate a brand kit for Kuro, an AI coding agent for senior engineers."
Interpretation:
#0a0a0a#0a0a0a base / #1a1a1a surface / #f0ede8 text / #00d4aa teal accentnpm install kuro, browser chrome with app.kuro.devUser: "Brand kit for TrustPaw, a compliance monitoring tool with a friendly-but-serious tone."
Interpretation:
#fdfbf7#fdfbf7 base / #f5f0e8 surface / #1a1a1a text / #1e40af deep blue accent + #b91c1c crimson secondaryUser: "Brand board for LuckDrop, a mobile game with loot drops and competitive leagues."
Interpretation:
#050505 to dark purple #0d0015#050505 base / #0d0015 surface / #f0f0f0 text / #22c55e neon green accent + #eab308 gold secondaryUser: "Brand kit for Athena, an AI research assistant that synthesizes scientific papers."
Interpretation:
#faf5ef#faf5ef base / #f0e8da surface / #1a1a1a text / #b8a472 muted gold accent / #1a2744 deep navy secondaryUser: "Brand kit for Lumora, a creative branding agency."
Interpretation:
#ff4d00 as the board canvas#ff4d00 primary canvas / #6f2dbd secondary / #1a1a4e dark surface / #fafafa textUser: "Brand kit for Thesis, a design feedback tool. Calm, soft gradients, minimal."
Interpretation:
MAmattpocock/skills288k近 7 天新增 10,178 个 star
AFaffaan-m/ECC278k近 7 天新增 4,046 个 star
GAgarrytan/gstack136k近 7 天新增 883 个 star
NEnextlevelbuilder/ui-ux-pro-max-skill135k近 7 天新增 1,864 个 star
NEnextlevelbuilder/ui-ux-pro-max-skill135k近 7 天新增 1,864 个 star
NEnexu-io/open-design101k近 7 天新增 1,126 个 star