Iván

Design

設計創意v1.0.3
作者:Iván更新於 2026-07-26共 4 個版本 · 最新版本釋出於 2026-07-26design@clawhub_ivangdavila/design
影片處理視覺素材UI 設計

據 AlphaMoat 統計,截至 2026-10,Design 累計下載 9,663 次、獲 6 個星標,按累計下載量在本次參與排名的 10,914 個 AI 技能中排名第 2,839 位,在「設計創意」分類(1,116 個)中排名第 370 位。

列表預設按累計下載量排序,也可按星標排序;詳情頁排名按累計下載量計算。

累計下載
9663
星標
6
全站下載量排名
#2,839
全站前 27% · 「設計創意」前 34%
平台累計下載星標評論
ClawHub ↗64286-
SkillHub ↗96636—

各平台的計數口徑不同,統計時間不一致。

這個技能是做什麼的

它圍繞層級、間距、字號體系、顏色和佈局給出可操作規則,也覆蓋可訪問性、響應式介面、幻燈片、圖表、郵件與文件等場景。設計偏好可能存入本地持久化檔案;如不希望跨會話複用,應檢查或移除相關檔案。

資料釋出於 2026-10-27

可以用來做什麼

  • 診斷介面或視覺稿的層級與擁擠問題
  • 為介面、幻燈片、圖表或文件制定視覺規則
  • 檢查可訪問性和響應式佈局

什麼時候用

  • 需要設計或評審視覺內容時
  • 版面顯得雜亂、平淡或難以閱讀時
  • 選擇字型、顏色或適配移動端時

示例提示詞

以下是根據技能文件生成的示例,不代表實際執行結果。

請檢查這個介面的層級、間距和可讀性,並給出修改清單。
幫我為這份演示稿制定一致的字型和配色建議。
檢查這張圖表在顏色和標註上的可訪問性問題。

月度資料趨勢

累計下載9663
144058561萬2026-03 · 累計下載 20492026-05 · 累計下載 34462026-06 · 累計下載 38012026-07 · 累計下載 50862026-08 · 累計下載 55682026-09 · 累計下載 61542026-10 · 累計下載 966326-0326-0926-10
星標6
4562026-03 · 星標 42026-05 · 星標 52026-06 · 星標 42026-07 · 星標 62026-08 · 星標 62026-09 · 星標 62026-10 · 星標 626-0326-0926-10

2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。 該技能有同名的其他版本,調整前的資料歸屬當前下載量最大的這一個。

月份累計下載環比星標
2026-109663—6
2026-096154—6
2026-085568+9.5%6
2026-075086+33.8%6
2026-063801+10.3%4
2026-053446—5
2026-032049—4

SKILL.md 原文

以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。

14 KB · ClawHub v1.0.3
---
name: Design
slug: design
version: 1.0.3
changelog: "Display name shown correctly"
description: Executes and critiques visual design with quantified rules for hierarchy, spacing, type scale, color, and layout. Use when designing UI screens, landing pages, slides, posters, social graphics, emails, charts, or PDFs, when a design looks off, cluttered, amateur, or flat and the user cannot say why, when nothing stands out or text is hard to read, when choosing fonts, pairing typefaces, or building a palette without a design system, when adapting a screen to dark mode or mobile, or when reviewing generated HTML/CSS before delivery. Not for design tokens and component libraries (design-system) or brand strategy (branding).
homepage: https://clawic.com/skills/design
metadata:
  clawdbot:
    emoji: "🎨"
展開全部(共 152 行)
    displayName: Design
    configPaths:
    - ~/Clawic/data/design/
---

User preferences and memory live in `~/Clawic/data/design/` (see `setup.md` on first use, `memory-template.md` for the file format). If you have data at an old location (`~/design/` or `~/clawic/design/`), move it to `~/Clawic/data/design/`.

## Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in `~/Clawic/data/design/config.yaml`.

| Variable | Type | Default | Effect |
|---|---|---|---|
| type_scale_ratio | number (1.125-1.5) | 1.25 | Regenerates the size ladder in Core Rule 4 (1.25 → 16/20/25/31/39); 1.333 for editorial and marketing pages |
| base_unit | 4 \| 8 (px) | 8 | Every spacing value becomes a multiple of this; 4 for dense pro tools, 8 everywhere else |
| accessibility_target | AA \| AAA | AA | AA floors are 4.5:1 body and 3:1 large text; AAA raises them to 7:1 and 4.5:1 across all checks |
| platform | web \| ios \| android \| print | web | Switches units (px/pt/dp), touch-target minimums (44pt HIG / 48dp Material), and platform conventions in `ui-screens.md` and `documents.md` |
| brand_file | path | none | Palette, typefaces, and logo rules read from `~/Clawic/data/design/brand.md`; overrides the derivation procedures in `palettes.md` and `fonts.md` |

Preference areas to record as the user reveals them:

- **output medium** — HTML/CSS vs image render vs written spec; framework in use (Tailwind, vanilla CSS) — affects what the deliverable looks like
- **aesthetic register** — minimal, dense, editorial, playful — shifts spacing generosity, type choices, and decoration tolerance
- **motion posture** — full motion vs reduced by default — gates Core Rule 8 and all transition suggestions
- **language and script** — RTL layouts mirror scan order; CJK needs looser line-height and no italic emphasis — affects `layout.md` and `fonts.md` guidance
- **critique style** — ranked punch-list vs annotated walkthrough — affects how `critique.md` findings are delivered

## When To Use

- Designing any visual artifact from scratch: UI screens, landing pages, slides, social graphics, posters, emails, PDFs.
- Fixing a design that "looks off" when the user cannot articulate why.
- Choosing typography, spacing, or a palette for a project that has no design system yet.
- Reviewing generated HTML/CSS or rendered images before delivery.
- Not for token pipelines and component libraries (design-system) or brand positioning (branding).

## Quick Reference

| Situation | Play |
|---|---|
| Design looks cluttered | Delete decoration first, then double the gap between groups; never add boxes to "organize" (→ `critique.md`) |
| "Looks off," cause unknown | Run gates in order: grayscale, squint, alignment count (→ Output Gates); full symptom chains in `critique.md` |
| Everything looks equally important | Rank all content 1-3; style only rank 1 as prominent, demote the rest to body style |
| Picking fonts | One family, weights 400/700; add a second family only to split display vs text roles (→ `fonts.md`) |
| Picking colors | 60-30-10 split; one accent hue; neutrals tinted with 2-5% of the accent's saturation (→ `palettes.md`) |
| Text hard to read | Check measure 45-75ch, line-height 1.5, contrast 4.5:1 before blaming the typeface |
| Mobile layout | 44pt/48dp touch targets, 16px minimum body (also prevents iOS input zoom), primary actions in bottom thumb zone |
| Slide deck | One idea per slide, 24pt text floor, max 6 lines of text per slide (→ `slides.md`) |
| Poster, thumbnail, social post | One-glance artifact: single message, readable at 10% zoom (→ `graphics.md`) |
| Form, table, or dashboard | Component-level rules in `ui-screens.md` |
| Chart or data display | Zero-baseline bars, direct labels, one hue for magnitude (→ `data-viz.md`) |
| HTML email | Tables and inline CSS, 600px body, Outlook renders with Word (→ `email.md`) |
| Dark mode requested | Not inversion: lighten surfaces to show elevation, desaturate accents 10-20%, white text at ~87% opacity (Material) (→ `dark-mode.md`) |
| Contrast, CVD, motion sensitivity | Floors and non-color encodings in `accessibility.md` |
| Print or PDF deliverable | 300 DPI, bleed, CMYK, margins (→ `documents.md`) |
| Any other visual task | Apply Core Rules in order 1-7: hierarchy, then spacing, then type, then color |

Depth on demand: `critique.md` symptom→cause when it looks off · `layout.md` grids, alignment, responsive · `fonts.md` choosing, pairing, setting type · `palettes.md` color without a system · `ui-screens.md` forms, tables, dashboards, states · `landing-pages.md` marketing pages and CTAs · `slides.md` decks · `graphics.md` posters, social, thumbnails · `documents.md` PDFs, reports, print · `email.md` HTML email · `data-viz.md` charts · `dark-mode.md` theming · `accessibility.md` contrast, CVD, motion, focus.

## Core Rules

1. **Hierarchy is a ranking, not a styling pass.** Inventory every element, rank each 1-3 by user importance (1 = must see, 2 = supporting, 3 = metadata). Exactly one rank-1 per view. Check: a grayscale screenshot must reveal the ranking in 3 seconds.
2. **Spend one channel per level step.** Adjacent hierarchy levels differ by one full step in ONE channel: size x1.25, weight +300 (400 to 700), or a clear lightness jump. Three small tweaks read as noise. Example: 18px/600 next to 16px/400 body is two half-steps and ranks poorly; use 20px/400 or 16px/700 instead.
3. **Spacing: base 8, factor 2 between grouping levels.** Scale 4/8/16/24/32/48/64 (multiples of `base_unit`). Between-group gap >= 2x within-group gap: 8px label-to-field, 24px field-to-field, 48px section-to-section. Proximity groups content more cheaply than any border or background box.
4. **Type floors: 16px body, 1.5 line-height, 45-75ch measure.** Bringhurst's 66ch is the target; under 45ch the eye jumps lines, over 75ch it loses the return sweep. Headings tighten to 1.1-1.3. Ratio `type_scale_ratio` = 1.25 gives 16/20/25/31/39 and covers most product work; 1.333 for editorial and marketing pages.
5. **Color: 60-30-10 with contrast floors.** ~60% neutral surface, ~30% secondary, ~10% accent, and the accent marks interactive or state-bearing elements only. WCAG floors: 4.5:1 body text, 3:1 large text (24px+, or 19px bold) and UI component boundaries. Never #000 on #FFF; near-black (#111 to #1a1a1a) cuts edge vibration.
6. **Every edge answers to a line.** Target <= 3 distinct vertical alignment lines per view; each extra line adds perceived clutter. Center only display text of 1-3 lines; left-align everything longer, because ragged-right gives the eye a fixed return point.
7. **Design in grayscale first.** Layout, hierarchy, and spacing must work with zero color. Add color last to encode meaning (state, action, brand); color used to rescue hierarchy fails for the ~8% of men with color vision deficiency and fails the grayscale gate.
8. **Motion: 150-300ms for UI transitions.** Ease-out on entry, ease-in on exit. Over 400ms reads as lag; under 100ms reads as a glitch. Animate only transform and opacity for smooth playback; honor reduced-motion preferences (`accessibility.md`).

## Hierarchy Procedure

1. Content inventory before any styling: list every element the view must contain, including legal text and empty states.
2. Rank each element 1-3. If two elements demand rank 1, split the view or demote one by position (top-left beats bottom-right in LTR scan order).
3. Assign styles top-down from the type scale. Demote rank 3 by lightness (40-60% gray on white), not by shrinking below 13px, where legibility collapses.
4. Place rank 1 in the natural entry point: top-left for documents, visual center for posters and slides.
5. Verify with the Output Gates before delivering.

## Spacing and Layout

- Macro before micro: set page margins and section gaps before touching padding inside components.
- Text content containers: 560-600px max width (keeps measure at 70-75ch at 16px, inside the 45-75ch range). Dashboards run full-width with 24px outer margins.
- Space belongs to what it labels: a heading sits closer to the text below it than to the section above (e.g. 8px below, 32px above). Equal gaps destroy grouping.
- Optical beats mathematical: icons, triangles, and play buttons need a 1-3px shift toward their visual weight to look centered. Grid math and column layouts: `layout.md`.
- Whitespace is the grouping mechanism. Filling "empty" corners with decoration deletes the structure the space was providing.

## Type and Color Execution

- Pair typefaces by role contrast (display serif + text sans), never by similarity; two near-identical sans faces read as a rendering mistake. Selection tables and loading: `fonts.md`.
- Skip adjacent weights: 400 vs 500 is indistinguishable at body sizes. Use 400/600 or 400/700.
- ALL CAPS only for labels of 1-2 words, with +5-10% letterspacing. Never track lowercase body text; it breaks the word shapes readers scan by.
- Palette derivation: pick the accent hue, build a 9-step lightness ramp from it, then tint neutrals with 2-5% of the accent's saturation. Pure grays next to a saturated accent look dirty. Full procedure: `palettes.md`.
- Semantic color is reserved: red for destructive and error, green for success. A red brand needs error states distinguished by a different shade plus an icon, never by hue alone.
- One radius scale per artifact (e.g. 4/8/16). Mixed radii on sibling elements is the fastest tell of unreviewed generated UI.

## Output Gates

Run before delivering any visual artifact:

- Grayscale test: desaturated, is the 1-3 ranking still obvious in 3 seconds?
- Squint test: blurred, do exactly the title and the primary action pop?
- Alignment count: <= 3 vertical alignment lines, and every element sits on one?
- Spacing audit: every gap a multiple of the base unit, between-group >= 2x within-group?
- Contrast check: every text/background pair meets its floor (4.5:1 body, 3:1 large/UI)?
- Single rank-1: exactly one primary action or focal point per view?

## Traps

| Trap | Why it fails | Do instead |
|---|---|---|
| Boxes and borders to separate content | Each border is added noise; proximity groups for free | Remove the box, double the between-group gap |
| Shrinking text to fit the space | Below 13px legibility collapses; the content problem remains | Cut content, paginate, or widen the container |
| Centering long text | No fixed return point for the eye, each line restart costs a search | Left-align; center only 1-3 line display text |
| Emphasizing with bold + bigger + colored at once | Burns all channels on one level; nothing is left to mark the true top rank | One channel per level step (→ Core Rule 2) |
| Color as the only state signal | Fails grayscale, print, and ~8% of male users | Encode with icon, weight, or position first; color reinforces |
| Pure gray neutrals beside a saturated accent | Grays pick up a dirty, mismatched cast by contrast | Tint neutrals with 2-5% of the accent hue |
| Filling every empty area | Whitespace was doing the grouping; filling it flattens hierarchy | Leave it; if it feels empty, the content ranking is wrong, not the space |
| Restyling before ranking | Polishing elements in place cements a flat hierarchy | Content inventory and 1-3 ranking first (→ Hierarchy Procedure) |
| Designing only the happy state | Empty, loading, error, and overflow states ship broken | Design the four non-happy states before polishing the happy one (`ui-screens.md`) |
| Checking one theme only | A palette tuned in light mode breaks contrast in dark | Run the Output Gates once per theme (`dark-mode.md`) |

## Where Experts Disagree

- **Pure black (#000) dark backgrounds.** OLED-first designers use #000 for power and contrast; Material argues #121212 reduces halation and lets shadows exist. Default #121212; go #000 only when the artifact targets OLED media consumption (video, photo feeds).
- **Grid math vs optical judgment.** Systematists snap everything to the 8px grid; typographers override math wherever the eye disagrees (icon centering, display-type spacing). Default: grid for layout, optical corrections allowed at component level — never the reverse.
- **System fonts vs webfonts.** System stacks cost zero load time and feel native; a webfont earns its bytes only when type IS the brand voice (marketing, editorial). Product UI defaults to a system stack or one variable font.
- **Minimal vs dense.** Consumer surfaces reward whitespace; daily-use pro tools reward density (more data per scroll, `base_unit` 4). The boundary is usage frequency, not taste: daily expert use → dense wins.

## Related Skills
More Clawic skills, get them at https://clawic.com/skills/design (install if the user confirms):
- `ui` — interface component patterns, interaction states, platform conventions
- `typography` — deep type work: optical sizes, variable weight axes, print settings
- `color` — full color systems, tokens, and cross-surface color validation
- `design-system` — scaling one-off decisions into tokens and component libraries
- `accessibility-audit` — full WCAG audit with screen-reader passes and remediation report

## Feedback

- If useful, star it: https://clawic.com/skills/design
- Latest version: https://clawic.com/skills/design

Part of [Clawic](https://clawic.com), the verified skill library. Get this skill: https://clawic.com/skills/design.

「設計創意」分類下的熱門技能

影片短影片圖片音樂音訊配音口播數位人全能創作設計創意

透過同一個AI創作工具完成AI圖片、影片、音樂與語音創作,查詢公開社交資料,編輯視覺成果,並集中檢視和管理生成資產。Beatra 全能創作支援AI圖片生成(文生圖)、AI影片生成(圖生影片)和AI影片編輯,透過AI音樂生成把想法做成歌曲或純音樂,也可用AI語音生成(文字轉語音)、多語言旁白和AI語音克隆完成跨媒介內容;還能查詢抖音資料、小紅書資料,以及 YouTube、TikTok、Instagram 與 X 上的公開帖子、筆記、評論、賬號和趨勢,把社交媒體資料用於競品調研和創作。它是一款面向AI內容創作與多模態內容創作的AI創作套件,用起來就是一個把圖片、影片、音樂與語音收在同一處的AI創作助手,適用於社交內容、產品展示、行銷素材、課程、播客、短片和跨媒介專案;還能在不同媒介間複用原始素材、已有成果和查到的公開資料,檢視製作進度、返回檔案、實際用量和積分記錄。

↓ 147.6萬★ 113u_319b171b
AI數位人口播與虛擬主播影片設計創意

用一張人像和一段短指令碼或語音,製作有明確講述內容的數位人口播影片。可使用現成錄音,也可以先選擇音色,把指令碼生成旁白,讓畫面中的人物隨聲音完成穩定、自然的口播。適合 AI 主播、虛擬主播影片、數位人講解、讓照片說話、照片說話影片、產品講解影片、數位人培訓影片、公告和社交口播,並在交付後檢查人物一致性、聲音清晰度、口型同步和動作穩定性。一節課的講解也能做成課程口播。

↓ 50.8萬★ 39u_319b171b
架構圖一鍵生成設計創意

ContextWeave 架構圖一鍵生成

↓ 47萬★ 280user_bddf3fe6
海報設計工作室設計創意

將一段活動描述、一張產品照片或品牌參考圖,轉化為抓人眼球的活動海報、宣傳 banner 或社交媒體配圖。這款 AI海報設計與製作工具生成圖片豐富的海報,在海報排版上搭配清晰視覺層級、按品類匹配的風格和預留文字區域,覆蓋活動、音樂節、電影、產品釋出和促銷場景。你可以把它當作 AI宣傳單製作工具來製作促銷宣傳單,當作活動海報設計工具,當作電影海報生成器,當作音樂節海報設計工具,當作產品釋出會海報的 banner設計 AI,或當作平面設計海報工作室。從活動主題出發,用品牌風格參考圖變換產品或場景照片,或在已認可草稿上做定向精修,完成可印刷或可分享的成品。

↓ 30.5萬★ 28user_87b8e34f
自動化克隆數位人口播混剪影片設計創意

抖音短影片雙路徑生產體系(單 Skill 版)。Path A[付費·高質量]:策略大腦→(可選:素材混剪/AI高光提取)→指令碼最佳化→AI生圖/素材準備→配音渲染(百鍊/GPT-SoVITS/CosyVoice/VoxCPM2+剪映/Remotion/ffmpeg/數位人)。Path B[免費·動畫風]:採集熱點→指令碼→edge-tts免費配音+HyperFrames開源渲染→MP4,零雲費開箱快。觸發詞分層:①"抖音短影片/選題規劃/合規閘門/營運策略" ②"口播指令碼/AI生圖/AI配音/影片渲染/Remotion渲染" ③"蹭熱點/快速出片/免費做影片" ④"混剪/高光提取/素材粗剪/autoclip

↓ 29.7萬★ 79user_3b54b400
GPT Image 2.5生圖設計創意

GPT Image 2.5生圖是一款 AI 工具,用 GPT-Image-2.5 和 ChatGPT Images 2.5 出一張圖。GPT Image 2.5 Studio 支援 gpt-image-2.5-flare 與 gpt-image-2.5-sunburst,可文生圖或參考圖生圖,做中文海報、商品圖、資訊圖和社媒畫面。

↓ 27.1萬★ 27user_87b8e34f
MiniMax H3 AI影片創作設計創意

用 MiniMax H3 這款 AI影片生成器創作高質感 2K AI 影片。從文字創意出發完成文生影片,讓靜態圖片自然動起來,用首尾畫面做首尾幀影片的完整轉場,也可以結合圖片、影片和音訊參考,以多模態影片生成塑造統一場景。適合 AI 廣告影片、電商產品影片、品牌宣傳影片、動態海報、遊戲 UI、電影片頭和社交短影片創作,形成電影感畫面與同步的原生立體聲。使用 Beatra 賬號即可開始,創作進度與成片集中管理。

↓ 26.3萬★ 23user_87b8e34f
AI Logo設計·品牌標識創作設計創意

輸入品牌名稱、行業方向或參考圖,生成專業 AI Logo、品牌標識和應用圖示。這款 AI logo 設計工具從你的品牌簡報出發——公司名、行業氣質、配色偏好——先探索多個商標設計方向,再用精確品牌色、簡潔幾何和可縮放構圖精修最強概念。製作企業 logo、創業公司標誌、字母標、徽章、組合標識和 VI設計中的主標,從網站 favicon 到門頭招牌都清晰可辨。從文字描述出發,將已有草圖和品牌素材轉化為精緻標識,或在已認可草稿上做定向精修。

↓ 26萬★ 19user_87b8e34f

Iván 的其他技能

資料月份:2026-10 · 下載/星標數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。