Imported from hardik140/youtube-automation (.agents/skills/hardik-thumbnail-creation/SKILL.md). Install upstream with npx skills add hardik140/youtube-automation --skill hardik-thumbnail-creation. Copyright stays with the author.

Thumbnail Creation Skill --- Hardik Personal Creator System

Purpose

This skill turns a video topic, script, reference thumbnails, and/or creator photos into a repeatable YouTube thumbnail production system.

The goal is not to make every thumbnail identical.

The goal is to preserve a recognizable Hardik creator identity while allowing the topic, emotion, clothing, environment, composition, and visual story to change from video to video.

Primary workflow:

Claude → Thumbnail Strategy → Nano Banana → Image Generation/Editing → Quality Check → Final Thumbnail

The system should remove repetitive decision-making and produce a strong first-pass thumbnail without requiring the same manual instructions every time.

1. CREATOR IDENTITY

1.1 Visual Identity

When Hardik appears in a thumbnail, maintain recognizable identity from the supplied reference photographs.

Reference characteristics:

• Young Indian male creator.
• Short, naturally textured/curly black hair.
• Light-to-medium beard and moustache.
• Defined eyebrows.
• Sharp, expressive eyes.
• Clean, masculine facial structure.
• Natural Indian appearance.
• Usually photographed from medium shot to upper-body/full-body depending on the concept.
• Can wear glasses/sunglasses when appropriate.
• Facial identity must remain consistent across thumbnails.
• Do not unnecessarily alter face shape, hairstyle, beard pattern, skin tone, or age.
• Avoid making him look like a generic AI-generated model.

1.2 Personality / On-Camera Presence

Hardik's thumbnail persona should feel:

• Curious
• Confident
• Intelligent
• Investigative
• Serious when the subject is serious
• Surprised when the subject requires discovery
• Skeptical when exposing misinformation
• Energetic when the subject is exciting
• Relatable rather than overly polished
• Young entrepreneur/creator/researcher rather than a traditional news anchor

The face should communicate the video's emotional promise before the viewer reads the title.

2. IMPORTANT: CLOTHING MUST NOT BECOME STAGNANT

Clothing is not part of Hardik's permanent identity.

The face/person identity stays consistent; wardrobe changes according to the story.

Do NOT force the grey blazer, white shirt, black trousers, or sunglasses into every thumbnail.

Wardrobe logic

Choose clothing based on:

1. Topic
2. Emotional tone
3. Background
4. Color contrast
5. Story setting
6. Creator persona
7. Thumbnail readability

Possible wardrobe categories:

Investigative / Scam / Exposé

• Dark shirt
• Black/charcoal jacket
• Neutral overshirt
• Casual professional outfit
• Optional glasses

Business / Startup / Money

• Smart casual shirt
• Polo
• Blazer
• Clean premium outfit

Technology / AI

• Black/dark T-shirt
• Minimal modern shirt
• Casual tech-creator appearance
• Hoodie when appropriate

Travel / Adventure

• Casual T-shirt
• Overshirt
• Jacket
• Outdoor clothing

History / Mystery

• Neutral shirt
• Dark jacket
• Period-neutral casual styling

Serious Social Issues

• Simple dark or neutral T-shirt/shirt
• Minimal accessories
• Avoid fashion-forward clothing that distracts from the subject

High-Energy / Viral Challenge

• Bright or contrasting casual outfit
• T-shirt
• Jacket/overshirt
• More dynamic pose

Premium / Success / Business

• Blazer
• Shirt
• Smart casual clothing
• Sunglasses only when visually useful

Wardrobe rule

Never say "always wear the grey blazer."

Instead:

Preserve Hardik's face and identity, but select clothing appropriate to the video's narrative and visual palette.

Wardrobe variation should make the channel feel alive.

3. THUMBNAIL CORE PRINCIPLE

A thumbnail is NOT a miniature summary of the video.

It is a visual curiosity engine.

The thumbnail should answer:

"Why should I click this?"

without answering:

"What happens in the entire video?"

The strongest thumbnail usually contains:

ONE dominant idea + ONE dominant emotion + ONE unresolved question.

4. THUMBNAIL PSYCHOLOGY

Use these principles when generating concepts.

4.1 Curiosity Gap

Show enough information to understand the subject but not enough to resolve it.

Weak:

"How This Scam Works"

Stronger visual concept:

Hardik holding the suspicious product + shocked/skeptical face + red warning symbol + one unexplained detail.

The viewer thinks:

"What exactly is wrong with this?"

4.2 Pattern Interrupt

Create something visually unexpected.

Examples:

• Ordinary object with shocking context.
• Huge scale difference.
• Before/after contrast.
• Human face reacting to something bizarre.
• Familiar place shown in an unexpected state.
• Normal product surrounded by warning signals.

4.3 Emotional Face

The face is a visual amplifier.

Choose expression deliberately:

Shock

• Raised eyebrows
• Open mouth
• Wide eyes
• Slight head movement

Suspicion

• Narrowed eyes
• Slight eyebrow tension
• Closed/partially closed mouth

Anger

• Lowered eyebrows
• Tight jaw
• Direct gaze

Fear

• Wide eyes
• Raised eyebrows
• Slightly open mouth

Confusion

• One eyebrow slightly raised
• Head tilted
• Tight/lowered mouth

Confidence

• Direct eye contact
• Controlled expression
• Slight smile

Sadness / Seriousness

• Reduced smile
• Softer eyes
• Neutral mouth

Do not use exaggerated emotion when the subject does not support it.

5. THUMBNAIL VISUAL HIERARCHY

Every thumbnail should have a hierarchy.

Priority 1 --- Main subject

The viewer should immediately recognize:

• Hardik
• Product
• Person
• Location
• Event
• Object

Priority 2 --- Story conflict

Examples:

• Scam
• Mystery
• Danger
• Contradiction
• Before/after
• Hidden truth
• Money
• Failure
• Discovery

Priority 3 --- Text

Text supports the image.

Text should NOT explain everything.

6. TEXT RULES

Ideal text

Usually:

2--5 words

Examples:

• "THE REAL TRUTH"
• "WAS IT FAKE?"
• "₹5 CR LOSS"
• "THEY LIED"
• "BANNED?"
• "WHY?"
• "BIGGEST SCAM?"
• "WHAT HAPPENED?"
• "THE DARK SIDE"
• "NO ONE KNEW"

Text hierarchy

Use:

• One dominant word
• One supporting phrase if necessary
• Strong contrast
• Large typography
• Minimal clutter

Avoid:

• Full sentences
• Paragraphs
• Excessive labels
• Small text
• Too many fonts
• Too many colors
• Repeating the YouTube title word-for-word

7. TEXT SHOULD CREATE A QUESTION

Whenever possible, text should generate an unresolved thought.

Examples:

Topic:

A suspicious product

Text: "100% PURE?"

Topic:

Historical mystery

Text: "WAS IT FAKE?"

Topic:

Financial loss

Text: "₹5 CR LOSS"

Topic:

Government controversy

Text: "THE REAL STORY"

The visual should provide context; the text should provide tension.

8. COMPOSITION SYSTEM

Default canvas:

16:9 YouTube thumbnail

Recommended working size:

1280 × 720

Use safe margins around all important text and faces.

Common compositions

Composition A --- Creator + Evidence

Left: - Hardik

Right: - Main object/evidence

Center: - Minimal text

Use for: - Investigations - Product analysis - Scams - Fact checking

Composition B --- Split Screen

Left: - Before / bad / problem

Right: - After / good / truth

Hardik can occupy one side or overlap the center.

Use for: - Comparisons - Countries - Social issues - Transformation

Composition C --- Creator + Huge Background Event

Foreground: - Large Hardik face/body

Background: - Major event/location

Use for: - Disasters - Politics - War/history - Major news - Mysteries

Composition D --- Object Hero

Large product/object in foreground.

Hardik reacts beside it.

Use for: - Products - Money - Technology - Food - Consumer investigations

Composition E --- Evidence Board

Use:

• Screenshot
• Document
• Photo
• Map
• Product
• Arrow
• Circle
• Warning symbol

But limit visual clutter.

9. SUBJECT SCALE

The thumbnail must survive mobile viewing.

Main face:

• Large enough to identify immediately.
• Prefer head/upper torso over tiny full-body character when emotion is important.

Main object:

• Large enough to recognize without zooming.

Background:

• Supporting role only.

Do not make everything equally large.

10. BACKGROUND SYSTEM

Background should communicate context immediately.

Possible backgrounds:

• Factory
• Street
• Laboratory
• Airport
• City
• Parliament/news environment
• Historical location
• Computer/technology environment
• Product shelf
• Disaster scene
• Map
• Document/evidence wall

Use depth:

Foreground → subject → evidence → atmospheric background

Background should be visually interesting but less dominant than the main story.

11. COLOR SYSTEM

Do not use one fixed palette for every thumbnail.

Use colors according to emotional category.

Danger / Scam / Exposé

• Red
• Black
• White

Mystery

• Black
• Deep blue
• White
• Red accent

Money

• Green
• Black
• Gold
• White

Technology

• Dark neutral
• Blue/cyan accents
• White

Disaster

• Dark blue
• Orange
• Red
• Smoke tones

Positive / Discovery

• Yellow
• White
• Blue
• Green

Travel

• Natural environment colors with strong subject separation

Color should create contrast, not decoration.

12. DEPTH AND LIGHTING

Use cinematic depth.

Foreground:

• Hardik
• Main object

Midground:

• Supporting evidence

Background:

• Context

Lighting:

• Strong subject separation
• Natural-looking highlights
• Controlled shadows
• Edge separation where necessary
• Slight cinematic contrast

Avoid:

• Plastic skin
• Excessive HDR
• Overly glossy AI skin
• Completely unrealistic lighting
• Flat studio lighting when the concept requires environmental lighting

13. CUTOUT STYLE

Hardik may be separated from the original photograph and composited into a new environment.

Maintain:

• Face proportions
• Hair
• Beard
• Skin texture
• Natural body proportions

Allowed:

• Pose changes
• Clothing changes
• Lighting changes
• Background replacement
• Controlled expression changes
• Camera-angle adaptation when plausible

Avoid identity drift.

14. POSE SYSTEM

Never reuse the exact same pose automatically.

Generate poses that communicate the story.

Examples:

• Pointing at evidence
• Holding an object
• Looking toward the evidence
• Looking directly at viewer
• Hands raised in surprise
• One hand explaining
• Arms crossed
• Holding document
• Leaning toward object
• Running/walking
• Looking behind
• Investigating
• Thinking
• Warning gesture

The pose should be causally connected to the story.

15. PROP SYSTEM

Props should increase understanding.

Useful props:

• Product
• Phone
• Laptop
• Document
• Money
• Map
• Screenshot
• Sign
• Warning icon
• Newspaper
• Photograph
• Tool
• Food
• Vehicle

Avoid random props.

Every prop should answer:

"Does this help the viewer understand the conflict?"

If not, remove it.

16. THUMBNAIL FORMULA LIBRARY

Formula 1 --- "What is wrong?"

Creator + suspicious object + warning

Formula 2 --- "They lied"

Creator + official-looking evidence + contradiction

Formula 3 --- "Before vs After"

Two contrasting environments + creator reaction

Formula 4 --- "Hidden truth"

Large subject + partially concealed evidence + question text

Formula 5 --- "Impossible"

Creator + impossible-looking visual + question

Formula 6 --- "Money"

Large ₹ figure + creator + object/company/event

Formula 7 --- "Danger"

Creator + dangerous scene + visual warning

Formula 8 --- "Mystery"

Dark environment + unknown object/event + creator looking toward it

Formula 9 --- "Exposé"

Creator + product/company + evidence + red stamp/marker

Formula 10 --- "Human story"

Creator + emotionally powerful person/event + minimal text

17. REFERENCE THUMBNAIL ANALYSIS

When reference thumbnails are provided, Claude should NOT copy them literally.

Instead extract:

1. Composition
2. Subject count
3. Subject scale
4. Facial emotion
5. Text length
6. Text placement
7. Typography hierarchy
8. Color contrast
9. Background complexity
10. Object placement
11. Use of arrows/circles
12. Lighting
13. Curiosity mechanism
14. Visual conflict
15. Why the thumbnail can work at mobile size

Then create a new composition using the same underlying principles.

18. CLAUDE'S ROLE

Claude is the thumbnail strategist and art director.

Claude should receive:

• Video title
• Script
• Topic
• Key claim
• Emotional angle
• Target audience
• Reference thumbnails
• Hardik reference photos
• Optional previous thumbnail

Claude should produce:

A. Thumbnail promise

One sentence explaining why someone would click.

B. Core emotion

Choose one:

• Shock
• Curiosity
• Fear
• Anger
• Suspicion
• Surprise
• Awe
• Sadness
• Confidence

C. Main visual conflict

What two things visually collide?

D. Hero subject

Who/what dominates the image?

E. Supporting evidence

What proves or hints at the story?

F. Text

Provide 3--5 short text options.

G. Composition

Explain left/right/center placement.

H. Expression

Specify Hardik's expression.

I. Pose

Specify the pose.

J. Wardrobe

Choose clothing appropriate to the topic.

K. Background

Specify environment.

L. Color direction

Specify contrast strategy.

M. Nano Banana prompt

Generate one production-ready prompt.

N. Negative constraints

List what must not happen.

19. NANO BANANA PROMPT ARCHITECTURE

Use this structure:

Create a high-impact YouTube thumbnail in 16:9 format for a video about [TOPIC].

Main subject: Hardik, using the supplied reference photos as the identity reference. Preserve his recognizable facial identity, hairstyle, beard, skin tone, facial proportions and overall appearance.

Expression: [EXPRESSION].

Pose: [POSE].

Wardrobe: [TOPIC-APPROPRIATE CLOTHING]. Do not force a fixed outfit; wardrobe should naturally fit the story.

Main visual: [MAIN OBJECT/EVENT].

Supporting visual: [EVIDENCE/SECONDARY ELEMENT].

Composition: [LAYOUT].

Background: [BACKGROUND].

Lighting: cinematic, realistic, strong subject separation, natural skin texture, believable environmental lighting.

Color direction: [COLOR PALETTE].

Typography: [TEXT], large bold highly readable typography with strong contrast and clean hierarchy.

Emotion: [EMOTIONAL PROMISE].

The thumbnail should communicate [CORE QUESTION] within one second.

Make the composition visually simple, dramatic, readable at mobile size, and optimized for high click-through potential.

Do not overcrowd the image.

Do not create unnecessary objects.

Do not make Hardik look like a different person.

Do not make the clothing identical to previous thumbnails unless specifically requested.

Do not use generic AI-model facial features.

Do not use excessive text.

Do not make the background compete with the hero subject.

20. IDENTITY CONSISTENCY VS VISUAL VARIATION

This is the most important rule.

Keep constant:

• Face identity
• Hair identity
• Beard identity
• General age
• Skin characteristics
• Creator personality
• Recognizable facial proportions

Vary:

• Clothing
• Pose
• Expression
• Camera angle
• Background
• Props
• Lighting
• Color palette
• Composition
• Accessories
• Body crop
• Story environment

Think:

Same creator --- different story.

Not:

Same creator --- same thumbnail template.

21. THUMBNAIL QUALITY CHECK

Before approving a thumbnail, ask:

1-second test

Can I understand the general story in one second?

Mobile test

Would this still work as a small YouTube thumbnail?

Face test

Is Hardik recognizable?

Emotion test

Does the expression communicate something?

Curiosity test

Does the image make me ask a question?

Contrast test

Can I distinguish foreground from background?

Text test

Can I read the main text instantly?

Simplicity test

Is there one dominant idea?

Originality test

Does it feel like a new thumbnail rather than a copy?

Wardrobe test

Does the clothing fit the topic and differ naturally from previous thumbnails?

AI artifact test

Are hands, eyes, teeth, glasses, text and objects believable?

If any major test fails, regenerate or edit.

22. AVOID THESE COMMON FAILURES

Never automatically produce:

• Tiny creator
• Tiny text
• Too many people
• Too many objects
• Full paragraph text
• Generic stock-photo composition
• Flat background
• Emotionless face
• Repeated grey blazer
• Repeated exact pose
• Repeated red/black layout for every topic
• Excessive arrows
• Excessive circles
• Random emojis
• Overly fake facial expressions
• Plastic skin
• Deformed hands
• Broken typography
• Unreadable screenshots
• Background clutter
• Multiple competing focal points

23. AUTOMATION PIPELINE

INPUT

User provides:

TITLE:
[YouTube title]

TOPIC:
[Topic]

SCRIPT:
[Full script or summary]

KEY CLAIM:
[Most important discovery]

REFERENCE THUMBNAILS:
[Images]

CREATOR REFERENCES:
[Hardik photos]

CLAUDE PROCESS

Claude automatically:

1. Understands the video.
2. Finds the strongest curiosity angle.
3. Identifies emotional promise.
4. Identifies the main visual conflict.
5. Generates 3 thumbnail concepts.
6. Scores concepts.
7. Selects the strongest concept.
8. Selects appropriate wardrobe.
9. Designs composition.
10. Creates Nano Banana prompt.
11. Creates negative constraints.
12. Generates final QA checklist.

NANO BANANA PROCESS

1. Use Hardik reference images.
2. Preserve identity.
3. Apply selected expression.
4. Apply selected pose.
5. Apply topic-specific wardrobe.
6. Build environment.
7. Add main object/evidence.
8. Compose for 16:9.
9. Generate.
10. Inspect.
11. Correct identity/artifacts/text.
12. Produce final candidate.

24. THREE-CONCEPT SYSTEM

Claude should normally generate three concepts.

Concept A --- Safe / Proven

Strong conventional YouTube composition.

Concept B --- High Curiosity

More unusual visual contradiction.

Concept C --- High Emotion

Maximum emotional impact.

Then score:

Factor Weight

Curiosity 25% Instant comprehension 20% Emotional impact 20% Mobile readability 15% Visual contrast 10% Originality 10%

Select the highest scoring concept.

25. TITLE + THUMBNAIL RELATIONSHIP

The title and thumbnail should work together.

Do not repeat the same information.

Example:

Title

"How This ₹10 Product Became a Massive Scam"

Thumbnail

"THEY LIED?"

Visual: Hardik + product + suspicious evidence.

The title provides context.

The thumbnail provides emotion and curiosity.

26. PERSONAL BRAND RULE

Hardik's thumbnails should gradually become recognizable as a creator brand, not merely individual viral images.

Recognition should come from:

• Face
• Expression style
• Strong visual storytelling
• Investigative/curious personality
• Clean composition
• Bold typography
• High contrast
• Consistent quality

Not from:

• One fixed outfit
• One fixed background
• One fixed color
• One fixed pose
• One fixed thumbnail template

27. FINAL MASTER INSTRUCTION FOR CLAUDE

When asked to create a thumbnail:

Act as a senior YouTube thumbnail strategist, visual storyteller and creative director.

Analyze the video's topic, title, script and supplied thumbnail references.

Do not simply reproduce the reference thumbnails. Extract the psychological and visual principles that make them effective.

Build a thumbnail around one dominant idea, one dominant emotion and one unresolved curiosity gap.

When Hardik appears, use the supplied creator photos as identity references. Preserve his recognizable facial identity, hairstyle, beard, skin tone and natural facial proportions.

Do NOT lock Hardik into a fixed wardrobe. Clothing must change naturally according to the topic, environment, emotional tone and composition. The grey blazer outfit is one possible wardrobe reference, not a permanent uniform.

Vary pose, expression, camera angle, clothing, background, lighting and composition while keeping creator identity consistent.

Prioritize mobile readability, strong subject separation, high emotional clarity, visual contrast and curiosity.

Avoid clutter, excessive text, generic AI faces, repetitive layouts, repetitive clothing and unnecessary decorative elements.

First generate three thumbnail concepts:

1. Proven / safe
2. High curiosity
3. High emotion

Score them using:

• Curiosity: 25%
• Instant comprehension: 20%
• Emotional impact: 20%
• Mobile readability: 15%
• Visual contrast: 10%
• Originality: 10%

Select the strongest concept.

Then provide:

1. Thumbnail concept
2. Emotional hook
3. Curiosity gap
4. Composition
5. Hardik expression
6. Hardik pose
7. Wardrobe
8. Background
9. Props/evidence
10. Text options
11. Color direction
12. Nano Banana production prompt
13. Negative constraints
14. Final QA checklist

The objective is not merely a beautiful image.

The objective is a thumbnail that makes a viewer stop scrolling and think:

"Wait... what is going on here?"

28. QUICK COMMANDS

These commands can be used with the skill.

/thumbnail

Create the complete thumbnail strategy.

/thumbnail3

Generate three distinct concepts.

/thumbnailhook

Find the strongest curiosity/emotional angle.

/thumbnailprompt

Convert the approved concept into a Nano Banana prompt.

/thumbnailfix

Analyze a generated thumbnail and list precise corrections.

/thumbnailidentity

Check Hardik's facial/visual identity consistency.

/thumbnailvariation

Create a new composition without changing the creator identity.

/thumbnailtext

Generate short thumbnail text options.

/thumbnailqa

Perform the final thumbnail quality audit.

29. GOLDEN RULE

Identity consistency ≠ visual repetition.

Hardik should be recognizable in every thumbnail.

But the viewer should never feel:

"I've already seen this exact thumbnail."

The system should produce:

A consistent creator + constantly changing visual story.

That is the foundation of the automated Hardik thumbnail system.