youtube-thumbnail-design

作者: halt-catch-fire

YouTube 縮圖設計,包含特定尺寸、對比規則與行動裝置預覽最佳化。涵蓋安全區域、文字擺放、臉部表情心理學及 A/B 測試。適用於:YouTube 縮圖、影片封面圖片、點擊率優化。觸發詞:youtube thumbnail、thumbnail design、video thumbnail、click through rate、ctr optimization、youtube cover、video cover image、thumbnail maker、thumbnail tips、youtube design、video preview image

npx skills add https://github.com/halt-catch-fire/skills --skill youtube-thumbnail-design

Install the belt CLI skill: npx skills add belt-sh/cli

YouTube Thumbnail Design

Create high-CTR YouTube thumbnails with AI image generation via inference.sh CLI.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

# Generate a thumbnail
belt app run falai/flux-dev-lora --input '{
  "prompt": "YouTube thumbnail style, close-up of a person with surprised excited expression looking at a glowing laptop screen, vibrant blue and orange color scheme, dramatic studio lighting, shallow depth of field, high contrast, cinematic",
  "width": 1280,
  "height": 720
}'

Specifications

SpecValue
Dimensions1280 x 720 px (minimum)
Recommended1920 x 1080 px
Aspect ratio16:9
Max file size2 MB
FormatsJPG, GIF, PNG

The 120px Test

Your thumbnail appears at roughly 120px wide on mobile — that's how most viewers first see it.

At 120px, viewers must be able to identify:

  1. The mood/emotion (from colors and expression)
  2. The general subject (from composition)
  3. The text (if any — only if large enough)

Test: view your thumbnail at 120px width. If it's a muddy blur, redesign.

Safe Zones

┌─────────────────────────────────────────────┐
│                                             │
│   ✅ SAFE FOR TEXT AND KEY ELEMENTS         │
│                                             │
│                                             │
│                                             │
│                                             │
│                                       ┌───┐ │
│                                       │ ⏱ │ │ ← Timestamp overlay
│                              ┌────────┴───┘ │    (bottom-right)
│   ┌────┐                     │  DURATION    │
│   │ CH │ Chapter marker      └──────────────│
└───┴────┴────────────────────────────────────┘
     ↑ Bottom-left: chapter/progress markers

Avoid placing critical elements in:

  • Bottom-right corner (video duration timestamp)
  • Bottom-left corner (chapter markers, progress bar)
  • Extreme edges (cropping varies by device)

Color Strategy

High-Contrast Pairs That Work

CombinationMoodBest For
Yellow + BlackUrgency, attentionTech, business, lists
Red + WhiteEnergy, excitementEntertainment, reactions
Blue + OrangeProfessional contrastEducation, tutorials
Green + WhiteGrowth, moneyFinance, success stories
Purple + YellowPremium, creativeDesign, art, creativity
White + DarkClean, minimalLuxury, minimalist channels

Color Rules

  • Background and text/subject should be complementary or high-contrast
  • Avoid same-temperature colors touching (red on orange = mud)
  • Use 3 colors maximum per thumbnail
  • Saturate more than real life — thumbnails compete with bright UI

Text on Thumbnails

When to Use Text

  • Lists/numbers: "7 Tips", "Top 10"
  • Strong opinions: "STOP Doing This"
  • Results: "$10K in 30 Days"
  • Comparisons: "vs" between two things

When NOT to Use Text

  • The video title already says it (redundant)
  • The emotion/visual tells the story
  • You can't make it large enough to read at 120px

Text Rules

RuleReason
Max 6 wordsReadability at thumbnail size
Min 60pt equivalentMust be legible at 120px width
Bold sans-serif fontThin fonts disappear at small sizes
Contrast stroke/shadowEnsures readability on any background
No small textIf it's not readable small, cut it

Face Expression Psychology

Thumbnails with faces get higher CTR than faceless thumbnails. Expression matters:

ExpressionCTR ImpactBest For
Surprise/shockHighestReaction, reveal, discovery content
CuriosityHighTutorial, how-to, tips
ExcitementHighUnboxing, reviews, announcements
Concern/worryMedium-highWarning, mistake, problem content
ConfidenceMediumExpert advice, authority content
NeutralLowestAvoid unless your brand is minimalist

Face Composition Rules

  • Face should fill 30-50% of the thumbnail
  • Eyes looking toward the text or subject (directs viewer attention)
  • Eyes looking at camera = connection. Eyes looking at object = curiosity.
  • Place face on one side (usually left), text or subject on the other
# Generate a face-forward thumbnail
belt app run falai/flux-dev-lora --input '{
  "prompt": "close-up portrait of a man with genuinely surprised expression, mouth slightly open, raised eyebrows, looking at camera, left side of frame, vibrant teal background, dramatic rim lighting, YouTube thumbnail style, high contrast, cinematic",
  "width": 1280,
  "height": 720
}'

# Generate a face-looking-at-subject thumbnail
belt app run bytedance/seedream-4-5 --input '{
  "prompt": "person looking amazed at a glowing holographic chart showing upward growth, dramatic blue and green lighting, right side profile view, dark background, tech aesthetic, high energy",
  "size": "2K"
}'

Thumbnail Patterns by Content Type

Tutorial / How-To

belt app run falai/flux-dev-lora --input '{
  "prompt": "overhead flat lay of organized workspace with laptop showing code editor, colorful sticky notes, coffee cup, clean bright background, professional setup, tutorial style composition, warm lighting",
  "width": 1280,
  "height": 720
}'

Before/After

belt app run falai/flux-dev-lora --input '{
  "prompt": "split composition, left side dark and messy disorganized desk, right side bright clean organized minimalist workspace, dramatic contrast between chaos and order, clear dividing line in center, high contrast",
  "width": 1280,
  "height": 720
}'

Product Review / Comparison

belt app run falai/flux-dev-lora --input '{
  "prompt": "two products facing each other with dramatic lighting and sparks between them, competition battle concept, dark background with colorful rim lighting, versus comparison style, high energy, product photography",
  "width": 1280,
  "height": 720
}'

Listicle / Number

belt app run falai/flux-dev-lora --input '{
  "prompt": "dynamic arrangement of 7 different colorful objects floating in space against dark gradient background, each item distinct and clearly separated, energetic composition, vibrant saturated colors, studio lighting",
  "width": 1280,
  "height": 720
}'

A/B Testing

Test one variable at a time:

VariableTest A vs B
Face vs No faceSame composition, with/without person
ExpressionSurprise vs curiosity
Color schemeWarm vs cool palette
Text vs No textWith/without text overlay
BackgroundBright vs dark
CompositionLeft-facing vs right-facing subject
# Generate variant A
belt app run falai/flux-dev-lora --input '{
  "prompt": "..., bright yellow background, ...",
  "width": 1280, "height": 720
}' --no-wait

# Generate variant B (same prompt, different background)
belt app run falai/flux-dev-lora --input '{
  "prompt": "..., dark navy background, ...",
  "width": 1280, "height": 720
}' --no-wait

Thumbnail Checklist

  • 1280x720 minimum (1920x1080 preferred)
  • Under 2MB file size
  • Passes the 120px squint test
  • No critical elements in bottom-right (timestamp) or bottom-left (chapter)
  • Max 3 colors, high contrast
  • Text (if any) is max 6 words, bold, with contrast stroke
  • Face expression matches content energy (if applicable)
  • Doesn't duplicate the video title
  • Stands out from surrounding thumbnails (check your niche)
  • Works on both light and dark YouTube backgrounds

Common Mistakes

MistakeProblemFix
Too much textUnreadable at thumbnail sizeMax 6 words or no text
Low contrastDisappears in the feedUse complementary colors
Cluttered compositionEye doesn't know where to lookOne focal point
Generic stock photo feelNo personality, gets skippedAuthentic expressions, unique angles
Tiny detailsLost at 120pxBold, simple shapes
Same style every videoViewer fatigueVary within brand guidelines
Misleading thumbnailKills trust, hurts retentionMatch the actual content

Related Skills

npx skills add inference-sh/skills@ai-image-generation
npx skills add inference-sh/skills@image-upscaling
npx skills add inference-sh/skills@prompt-engineering

Browse all apps: belt app list

來自 halt-catch-fire 的更多技能

ai-image-generation
halt-catch-fire
透過 inference.sh CLI 使用 GPT-Image-2、FLUX、Gemini、Grok、Seedream、Reve 及 50 多種模型生成 AI 圖像。模型包括:GPT-Image-2、FLUX Dev LoRA、FLUX.2 Klein LoRA、Gemini 3 Pro Image、Grok Imagine、Seedream 4.5、Reve、ImagineArt。功能:文字轉圖像、圖像轉圖像、修補、LoRA、圖像編輯、放大、文字渲染。適用於:AI 藝術、產品模型、概念藝術、社交媒體圖形、行銷視覺、插圖。觸發詞:flux、圖像生成、AI 圖像、文字轉...
creativemediaimage
ai-video-generation
halt-catch-fire
We need to translate the given text from English to Traditional Chinese. The instruction says to preserve product names, protocol names, URLs, numbers, and technical terms. So names like Google Veo, Seedance 2.0, HappyHorse, Wan, Grok, inference.sh CLI, Veo 3.1, etc. should remain as is. Also preserve terms like text-to-video, image-to-video, etc. but translate the surrounding Chinese text. The text inside <text> is a description of an agent skill. We need to output only the translated text, no extra labels. Let's translate paragraph by paragraph: "Generate AI videos with Google Veo, Seedance 2.0, HappyHorse, Wan, Grok and 40+ models via inference.sh CLI." -> "通過 inference.sh CLI 使用 Google Veo、Seedance 2.0、HappyHorse、Wan、Grok 及 40 多個模型生成 AI 影片。" "Models: Veo 3.1, Veo 3,
creativevideomedia
twitter-automation
halt-catch-fire
We need to translate the given text from English to Traditional Chinese. The instruction says to preserve product names, protocol names, URLs, numbers, and technical terms. The name "twitter-automation" is to be preserved but not included unless it appears in the source text. The source text does not include that name, so we don't add it. We translate only the text inside <text>. No extra labels or commentary. The text: "Automate Twitter/X with posting, engagement, and user management via inference.sh CLI. Apps: x/post-tweet, x/post-create (with media), x/post-like, x/post-retweet, x/dm-send, x/user-follow. Capabilities: post tweets, schedule content, like posts, retweet, send DMs, follow users, get profiles. Use for: social media automation, content scheduling, engagement bots, audience growth, X API. Triggers: twitter api, x api, tweet automation, post to twitter, twitter bot, social media automation, x..." We need to translate terms like "Automate",
marketingapicommunication
ai-avatar-video
halt-catch-fire
透過 inference.sh CLI 建立 AI 虛擬角色與說話頭像影片。推薦:P-Video-Avatar(最快、最便宜、內建 TTS)。另可選:OmniHuman、Fabric、PixVerse。音訊:Inworld TTS-2(支援 100 多種語言、角色情感引導)、ElevenLabs、Kokoro。功能:音訊驅動虛擬角色、文字轉虛擬角色、唇形同步影片、說話頭像生成、虛擬主持人、UGC 內容。適用於:AI 主持人、解說影片、虛擬網紅、配音、行銷影片、UGC 廣告、遊戲虛擬角色等。
videocreativemedia
agent-browser
halt-catch-fire
透過 inference.sh 為 AI 代理提供瀏覽器自動化功能。可導覽網頁、使用 @e 參考與元素互動、擷取螢幕截圖、錄製影片。功能包括:網頁抓取、表單填寫、點擊、打字、拖放、檔案上傳、執行 JavaScript。適用於:網頁自動化、資料擷取、測試、代理瀏覽、研究。觸發詞:瀏覽器、網頁自動化、抓取、導覽、點擊、填寫表單、螢幕截圖、瀏覽網頁、playwright、無頭瀏覽器、網頁代理、上網、錄製影片
browser-automationweb-scrapingtesting
web-search
halt-catch-fire
透過 inference.sh CLI 使用 Tavily 和 Exa 進行網路搜尋與內容擷取。應用:Tavily 搜尋、Tavily 擷取、Exa 搜尋、Exa 回答、Exa 擷取。功能:AI 驅動搜尋、內容擷取、直接回答、研究。用途:研究、RAG 管線、事實查核、內容彙整、代理程式。觸發詞:網路搜尋、tavily、exa、搜尋 API、內容擷取、研究、網際網路搜尋、AI 搜尋、搜尋助理、網頁抓取、rag、perplexity 替代方案
researchweb-scrapingapi
infsh-cli
halt-catch-fire
透過 inference.sh CLI 執行 250 多個 AI 應用程式,包括圖片生成、影片創作、大型語言模型、搜尋、3D、Twitter 自動化。模型包含:FLUX、Veo、Gemini、Grok、Claude、Seedance、OmniHuman、Tavily、Exa、OpenRouter 等眾多選擇。適用於執行 AI 應用程式、生成圖片/影片、呼叫大型語言模型、網路搜尋或自動化 Twitter 時使用。觸發詞:inference.sh、infsh、ai model、run ai、serverless ai、ai api、flux、veo、claude api、image generation、video generation、openrouter、tavily、exa search、twitter api、grok
developmentapicreative
landing-page-design
halt-catch-fire
著陸頁轉換優化,包含佈局規則、英雄區塊設計與CTA心理學。涵蓋首屏公式、社會證明放置、行動設計與F型閱讀模式。適用於:新創著陸頁、產品頁面、SaaS行銷、轉換優化。觸發詞:著陸頁、英雄區塊、首屏、轉換優化、著陸頁設計、CTA按鈕、英雄圖片、著陸頁佈局、SaaS著陸頁、產品頁面設計、轉換率、著陸頁...