Skip to main content

Product Requirements Document: Really American Thumbnail Generator

Date: December 1, 2025 Version: 1.0 Product Name: Really American Thumbnail Generator Target Users: Really American Media content producers (Kenny Hesse, Tony, and team)

Executive Summary​

The Really American Thumbnail Generator is an AI-powered tool designed to dramatically reduce thumbnail creation time from 30-45 minutes to under 5 minutes while maintaining the channel's high-performing visual style. The tool leverages video transcripts and brand templates to generate thumbnails that match the channel's proven "Sensationalist Political Commentary" aesthetic.

Problem Statement​

Current Pain Points​

  • Thumbnail creation takes 30-45 minutes per video
  • Manual Photoshop process requires multiple approval steps
  • A/B testing data is collected but not analyzed
  • Inconsistent application of successful design patterns
  • Bottleneck in the production pipeline affecting time-sensitive content

Impact​

  • Delayed publication of time-sensitive political content
  • Reduced competitive advantage against AI-generated content channels
  • Inefficient use of creative resources
  • Lost revenue opportunity from delayed uploads

Solution Overview​

An AI-powered thumbnail generator that:

  1. Analyzes video transcripts to select optimal templates
  2. Generates thumbnails using NanoBanana Pro API
  3. Enables iterative refinement through text and image feedback
  4. Maintains brand consistency while allowing creative flexibility

Core Features​

1. Intelligent Template Selection​

Description: System analyzes video transcript and automatically selects the most appropriate template based on content type.

User Flow:

  1. User pastes full video transcript
  2. System analyzes for conflict narratives, key figures, and story type
  3. System recommends optimal template (A or B)
  4. User can override if needed

Technical Requirements:

  • NLP analysis for story categorization
  • Pattern matching for identifying "conflict," "humiliation," or "bad news" narratives
  • Confidence scoring for template recommendation

2. Pre-Script Thumbnail Generation​

Description: Generate thumbnails before script completion using story context.

User Flow:

  1. User enters story context in natural language
  2. System reformats into optimized prompt
  3. Thumbnail generated based on context
  4. Can be refined after script completion

Input Example: "Pete Hegseth got roasted on Twitter after arguing with Mark Kelly about military service"

System Translation:

  • Template: Conflict & Humiliation
  • Target: Pete Hegseth (panicked expression)
  • Opponent: Mark Kelly (stern/mocking)
  • Headline: "HEGSETH GETS DESTROYED"

3. Post-Script Thumbnail Generation​

Description: Generate thumbnails using full video transcript.

User Flow:

  1. User pastes complete video script
  2. System extracts key figures, quotes, and narrative
  3. Suggests template and populates variables
  4. User can modify text/images before generation

4. Iterative Refinement​

Description: Modify generated thumbnails without starting over.

Capabilities:

  • Text feedback: "Make Trump look more panicked"
  • Image swapping: "Replace Biden with this image [upload]"
  • Element adjustment: "Add arrow pointing to document"
  • Text modification: "Change headline to 'IT'S OVER!'"

Technical Requirements:

  • Image segmentation for element replacement
  • Prompt modification system
  • Version history tracking

5. Template System​

Template A: "Conflict & Humiliation" (Split Screen)​

Use Case: Person vs. Person narratives

Variables:

  • {TARGET_NAME}: Subject being humiliated
  • {TARGET_IMAGE}: Image or auto-generated expression
  • {OPPONENT_NAME}: Person doing the humiliating
  • {OPPONENT_IMAGE}: Image or auto-generated expression
  • {BUBBLE_TEXT}: Speech bubble quote (3-4 words max)
  • {HEADLINE_TEXT}: Bottom banner text (4-6 words)
  • {ARROW_DIRECTION}: From opponent to target

Template B: "Bad News & Data" (Collage Style)​

Use Case: Polls, crashes, protests, general disaster narratives

Variables:

  • {TARGET_NAME}: Central figure reacting
  • {TARGET_IMAGE}: Panicked/shocked expression
  • {EVIDENCE_ELEMENTS}: Background elements (graphs, crowds, documents)
  • {HEADLINE_TEXT}: Quote-style headline
  • {INSET_IMAGE}: Corner proof element
  • {ARROW_TARGET}: What arrow points to

Template C: "Freeflow" (AI Creative)​

Use Case: When templates A/B don't fit

Inputs:

  • Story context or transcript
  • Brand guidelines constraints
  • Reference thumbnails (optional)

User Interface Requirements​

Main Screen​

┌─────────────────────────────────────┐
│  Really American Thumbnail Generator │
├─────────────────────────────────────┤
│                                     │
│  [Paste Video Transcript Here]      │
│  _________________________________  │
│  _________________________________  │
│  _________________________________  │
│                                     │
│  OR                                 │
│                                     │
│  [Describe Your Story]              │
│  _________________________________  │
│                                     │
│  [Generate Thumbnail]               │
└─────────────────────────────────────┘

Generation Screen​

┌─────────────────────────────────────┐
│  Template: Conflict & Humiliation   │
├─────────────────────────────────────┤
│                                     │
│  Target: [Donald Trump     ] 📷     │
│  Expression: [Panicked    ▼]        │
│                                     │
│  Opponent: [Mark Kelly    ] 📷      │
│  Expression: [Mocking     ▼]        │
│                                     │
│  Speech Bubble: [HE'S DONE!]        │
│  Headline: [TRUMP GETS ROASTED]     │
│                                     │
│  [Preview] [Generate] [Save]        │
└─────────────────────────────────────┘

Iteration Interface​

┌─────────────────────────────────────┐
│  Generated Thumbnail                │
├─────────────────────────────────────┤
│  [Generated Image Display]          │
│                                     │
│  Feedback:                          │
│  [Make Trump more orange]           │
│  [Add Biden laughing on right]      │
│                                     │
│  Upload New Element: [Choose File]  │
│                                     │
│  [Apply Changes] [Download] [New]   │
└─────────────────────────────────────┘

Brand Guidelines Integration​

Typography Rules​

  • Font: Impact or Bebas Neue Bold
  • Case: ALL CAPS for headlines
  • Colors:
    • Primary: Yellow #FFF200
    • Secondary: White #FFFFFF
    • Emphasis: Red #FF0000
  • Effects: 3-5px black stroke, hard drop shadow

Color Palette​

  • Alert Yellow: #FFFF00
  • Alarm Red: #EE2626
  • Democrat Blue: #003366
  • Background Grays: #2C2C2C to #4A4A4A

Composition Rules​

  • Text placement: Bottom 1/3
  • Two opposing faces required
  • One unflattering/panicked expression
  • Arrow element mandatory
  • High saturation on all elements

Technical Architecture​

API Integration​

// NanoBanana Pro Integration
// FOLLOW GOOGLE'S GUIDELINES!

### Data Flow
1. Input → Transcript/Context
2. Analysis → Template Selection
3. Variable Extraction → Auto-populate fields
4. User Modification → Manual overrides
5. Generation → NanoBanana API call
6. Iteration → Feedback loop
7. Export → Download/Upload to production

## Success Metrics

### Primary KPIs
- Thumbnail creation time: < 5 minutes (from current 30-45 min)
- First-attempt satisfaction rate: > 70%
- Iterations needed: < 3 per thumbnail

### Secondary KPIs
- Template accuracy: > 80% correct auto-selection
- CTR improvement: Maintain or exceed current rates
- User adoption: 100% team usage within 2 weeks

## MVP Scope (Phase 1)

### Included
- Templates A & B hardcoded
- Transcript analysis for template selection
- Basic variable extraction
- Single iteration capability
- Download functionality

### Not Included (Phase 2)
- A/B testing integration
- Historical performance analysis
- Batch generation
- Direct YouTube upload
- Custom template creation

## User Stories

### Story 1: Quick Thumbnail for Breaking News
**As** Kenny Hesse
**I want to** generate a thumbnail in under 5 minutes
**So that** I can publish time-sensitive content quickly

**Acceptance Criteria:**
- Can paste transcript and get thumbnail in < 5 min
- Template auto-selected correctly
- One-click download ready for upload

### Story 2: Iterate on Generated Thumbnail
**As** Tony (thumbnail creator)
**I want to** refine AI-generated thumbnails
**So that** I maintain creative control and brand consistency

**Acceptance Criteria:**
- Can modify text after generation
- Can swap images/faces
- Can adjust expressions and emotions

### Story 3: Pre-Script Thumbnail Creation
**As** a content producer
**I want to** create thumbnails before finishing my script
**So that** I can parallelize production tasks

**Acceptance Criteria:**
- Natural language input accepted
- Generates viable thumbnail from context
- Can update after script completion

## Risks & Mitigation

### Risk 1: NanoBanana API Quality
**Mitigation:** Implement fallback templates, maintain Photoshop backup process

### Risk 2: Template Limitations
**Mitigation:** Include freeflow option, plan for custom template addition

### Risk 3: Brand Consistency
**Mitigation:** Strict validation rules, approval workflow option

## Appendix

### A. Sample Prompts for NanoBanana

#### Template A Prompt Structure

Create a YouTube thumbnail with vertical split-screen composition. Style: Sensationalist political commentary, high contrast, photorealistic.

LEFT SIDE:

  • Subject: {TARGET_NAME}, professional photo
  • Expression: Extreme {TARGET_EMOTION}
  • Background: Dark blurred news studio
  • Lighting: Cool blue tint

RIGHT SIDE:

  • Subject: {OPPONENT_NAME}, professional photo
  • Expression: {OPPONENT_EMOTION}
  • Background: Bright contrasting setting
  • Lighting: Warm tones

OVERLAYS:

  • Curved yellow arrow (#FFF200) from right to left subject
  • White speech bubble, black outline, text: "{BUBBLE_TEXT}"
  • Bottom banner: Impact font, yellow (#FFF200), black stroke (5px)
  • Text: "{HEADLINE_TEXT}"

Technical: 16:9 aspect ratio, high saturation, dramatic lighting


### B. Brand Validation Checklist
- [ ] Yellow text uses #FFF200
- [ ] Text has 3-5px black stroke
- [ ] Expressions are exaggerated
- [ ] Arrow element present
- [ ] Two opposing subjects visible
- [ ] Text in bottom 1/3 of image
- [ ] ALL CAPS formatting
- [ ] High color saturation

### C. Performance Benchmarks
Based on channel analysis:
- Top thumbnails average 400K+ views
- Common elements: Split faces, yellow text, arrows
- Optimal headline length: 3-5 words
- Speech bubbles increase CTR by ~15%