Use when generating premium design reference images across three modes — brand identity boards, website section images, or mobile app screens — art-directed and implementation-friendly, not generic AI slop. Output is images only, never code. Triggers on "brand kit", "logo 设计", "品牌系统", "web design image", "landing page image", "网站设计图", "mobile design image", "app screen image", "移动端设计图", "手机界面图". Not for frontend code (use frontend-design) or image-to-code pipeline (use image-to-code).
Generate premium design reference images across three modes — brand identity, web sections, mobile
screens — that feel art-directed and implementation-friendly, not generic AI slop. Output is images
only, never code.
When to use
Brand identity boards, logo concepts, identity systems → brand mode
Website section reference images (one horizontal image per section) → web mode
Mobile app screen concepts and multi-screen flows → mobile mode
Not for: frontend code (use frontend-design); image-to-code pipeline (use image-to-code).
Tool note: use an available image-generation tool — do not bind to a specific product.
Steps
1. Pick the mode and load its workflow
Infer from the brief which mode fits, then load that mode's reference for the full workflow. Each
mode has its own section-count rule, composition logic, and anti-slop list.
brand — brand-kit overview boards, logo concepting, identity systems → references/brand-mode.md
web — one horizontal section image per section, premium art direction → references/web-mode.md
mobile — phone-mockup-framed app screens and multi-screen flows → references/mobile-mode.md
If the brief spans modes (e.g. a brand system plus web mockups), lead with the primary mode and
note the secondary. Do not mash modes into one frame.
2. Hold one art direction across the set
Whatever the mode, lock one visual direction — palette, typography, imagery treatment, corner
radius, density — and keep it consistent across every image in the set. Screen 3 must not drift
into a different product world than screen 1. The mode reference defines the mode-specific bible;
this step is the shared consistency spine across the whole set.
3. Ban AI slop (all modes)
The shared anti-slop ban list (banned brands, filler verbs, purple-blue gradients, glassmorphism,
blobs, fake KPI columns) lives in
${CLAUDE_PLUGIN_ROOT}/skills/03-design/frontend-design/references/anti-tells.md —
load it and apply it. Mode-specific tells to also ban: stock-template boards, phone-shaped
websites. If a render is weak — tiny text, unclear spacing, fake navigation, clutter, inconsistent
framing — regenerate it. Do not settle for the first mediocre render.
4. Match references without copying
If docs/design/references.md exists (produced by design-research), load it as the primary
reference source — it pairs real published designs with pre-extracted layout/grid/spacing/type/
density/accent analysis. If the user provides reference images directly, extract their layout rhythm, grid, spacing, type scale,
visual density, and accent-color logic. Do NOT copy exact logos, brand names, compositions,
slogans, or unique visual assets. References are quality training, not templates.
Output: design reference image(s). Not code.
Verify
Mode chosen decisively (brand / web / mobile) and the mode reference loaded
One art direction held across every image in the set (palette, type, radius, imagery treatment)
No AI slop — no purple-blue gradients, blobs, cloned blocks, fake brands, generic copy
Weak renders regenerated — no tiny text, fake navigation, or clutter settled for
User references matched for quality/rhythm only — no exact logos, names, or compositions copied
Output is images only — a developer or designer could use them as reference, never code
1---2name: imagegen3description: Use when generating premium design reference images across three modes — brand identity boards, website section images, or mobile app screens — art-directed and implementation-friendly, not generic AI slop. Output is images only, never code. Triggers on "brand kit", "logo 设计", "品牌系统", "web design image", "landing page image", "网站设计图", "mobile design image", "app screen image", "移动端设计图", "手机界面图". Not for frontend code (use frontend-design) or image-to-code pipeline (use image-to-code).4---56# Imagegen78Generate premium design reference images across three modes — brand identity, web sections, mobile9screens — that feel art-directed and implementation-friendly, not generic AI slop. Output is images10only, never code.1112## When to use1314- Brand identity boards, logo concepts, identity systems → **brand** mode15- Website section reference images (one horizontal image per section) → **web** mode16- Mobile app screen concepts and multi-screen flows → **mobile** mode17- Triggers on "brand kit", "logo 设计", "品牌系统", "web design image", "landing page image", "网站设计图", "mobile design image", "app screen image", "移动端设计图", "手机界面图"1819**Not for:** frontend code (use `frontend-design`); image-to-code pipeline (use `image-to-code`).2021**Tool note:** use an available image-generation tool — do not bind to a specific product.2223## Steps2425### 1. Pick the mode and load its workflow2627Infer from the brief which mode fits, then load that mode's reference for the full workflow. Each28mode has its own section-count rule, composition logic, and anti-slop list.2930- **brand** — brand-kit overview boards, logo concepting, identity systems → [references/brand-mode.md](references/brand-mode.md)31- **web** — one horizontal section image per section, premium art direction → [references/web-mode.md](references/web-mode.md)32- **mobile** — phone-mockup-framed app screens and multi-screen flows → [references/mobile-mode.md](references/mobile-mode.md)3334If the brief spans modes (e.g. a brand system plus web mockups), lead with the primary mode and35note the secondary. Do not mash modes into one frame.3637### 2. Hold one art direction across the set3839Whatever the mode, lock one visual direction — palette, typography, imagery treatment, corner40radius, density — and keep it consistent across every image in the set. Screen 3 must not drift41into a different product world than screen 1. The mode reference defines the mode-specific bible;42this step is the shared consistency spine across the whole set.4344### 3. Ban AI slop (all modes)4546The shared anti-slop ban list (banned brands, filler verbs, purple-blue gradients, glassmorphism,47blobs, fake KPI columns) lives in48[${CLAUDE_PLUGIN_ROOT}/skills/03-design/frontend-design/references/anti-tells.md](${CLAUDE_PLUGIN_ROOT}/skills/03-design/frontend-design/references/anti-tells.md) —49load it and apply it. Mode-specific tells to also ban: stock-template boards, phone-shaped50websites. If a render is weak — tiny text, unclear spacing, fake navigation, clutter, inconsistent51framing — regenerate it. Do not settle for the first mediocre render.5253### 4. Match references without copying5455If `docs/design/references.md` exists (produced by `design-research`), load it as the primary56reference source — it pairs real published designs with pre-extracted layout/grid/spacing/type/57density/accent analysis. If the user provides reference images directly, extract their layout rhythm, grid, spacing, type scale,58visual density, and accent-color logic. Do NOT copy exact logos, brand names, compositions,59slogans, or unique visual assets. References are quality training, not templates.6061**Output:** design reference image(s). Not code.6263## Verify6465- [ ] Mode chosen decisively (brand / web / mobile) and the mode reference loaded66- [ ] One art direction held across every image in the set (palette, type, radius, imagery treatment)67- [ ] No AI slop — no purple-blue gradients, blobs, cloned blocks, fake brands, generic copy68- [ ] Weak renders regenerated — no tiny text, fake navigation, or clutter settled for69- [ ] User references matched for quality/rhythm only — no exact logos, names, or compositions copied70- [ ] Output is images only — a developer or designer could use them as reference, never code7172## References7374- [${CLAUDE_PLUGIN_ROOT}/references/engineering-principles.md](${CLAUDE_PLUGIN_ROOT}/references/engineering-principles.md) — shared discipline (surface assumptions, push back, verify don't assume)75- [${CLAUDE_PLUGIN_ROOT}/references/design-principles.md](${CLAUDE_PLUGIN_ROOT}/references/design-principles.md) — design discipline (CRAP, hierarchy before decoration, design every state, accessibility non-optional, consistency from systems)76- [${CLAUDE_PLUGIN_ROOT}/references/product-icon.md](${CLAUDE_PLUGIN_ROOT}/references/product-icon.md) — product-icon execution criterion (study OSS icon systems; not simple shapes + simple colors)77- [references/brand-mode.md](references/brand-mode.md) — brand identity board workflow (strategy, visual modes, logo concepting, board composition)78- [references/web-mode.md](references/web-mode.md) — web section image workflow (section count, hero scale, 9-axis variation engine)79- [references/mobile-mode.md](references/mobile-mode.md) — mobile screen flow workflow (platform mode, design bible, phone mockup, safe areas)
Run npx skillmds@latest add int2t05/imagegen in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Use when generating premium design reference images across three modes — brand identity boards, website section images, or mobile app screens — art-directed and implementation-friendly, not generic AI slop. Output is images only, never code. Triggers on "brand kit", "logo 设计", "品牌系统", "web design image", "landing page image", "网站设计图", "mobile design image", "app screen image", "移动端设计图", "手机界面图". Not for frontend code (use frontend-design) or image-to-code pipeline (use image-to-code). It is listed under Design & Media on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
int2t05 (@int2t05) published this skill. Their other Agent Skills are listed on their SkillMD profile.