Character Reference Generation

AI Image Generation Best Practices

Flux 1.1 Pro Documentation Insights:

• image_prompt parameter = Flux Redux composition guidance (NOT character consistency)
• prompt_strength parameter does NOT exist in Flux 1.1 Pro API
• Character consistency achieved through detailed text descriptions
• Pose variation requires explicit pose descriptions in prompt

Prompt Structure:

{CHARACTER_DESC}, {attire}, {scenario}, {pose}, {environment}, {lighting}, {mood}, {STYLE_BASE}

Anti-Patterns:

• ❌ Using image reference + vague prompts = repetitive poses
• ❌ Assuming image_prompt provides character consistency
• ❌ Generic pose descriptions = same pose repeated

Correct Approach:

• ✅ Detailed character description in text
• ✅ Specific pose per scenario ("leaning forward analyzing data", "standing gesturing toward display")
• ✅ No image reference unless using Flux Redux for composition

Character-Agnostic Workflow

Input Requirements:

1. Character profile (characters/{name}.md)
2. Style guide (prompt templates)
3. Scenario list (appropriate to character's narrative role)
4. Output directory structure

Script Structure:

// Configuration
const CHARACTER = {
  name: "Character Name",
  description: "detailed appearance from profile",
  age: number
};

const STYLE = "aesthetic appropriate to narrative";

const SCENARIOS = [
  // Array of scenario objects with:
  // - id, title, scenario, attire, pose, environment, lighting, mood
];

const OUTPUT_DIR = `assets/character-reference/{character-slug}/{collection-name}`;

Character-Specific Adaptations

Alex (21, detective noir, professional):

• Corporate/tech/investigation scenarios
• Navy/charcoal suits, casual professional
• Monitor glow, noir shadows, dramatic lighting

Iris (13, Wonderland native, fantasy):

• Garden/magic/Wonderland scenarios
• Living plant clothing, iridescent features
• Soft pastels, magical glow, nature lighting

Maya (15-16, contemporary teenager):

• School/social/casual scenarios
• Contemporary teen fashion
• Natural daylight, casual environments

Mom/Bell-keeper/Shimmer:

• Role-appropriate scenarios
• Character-specific styling
• Context-matched environments

Implementation Steps

1. Extract Character Profile

// Read characters/{name}.md
// Parse: physical description, age, personality, role
// Build CHARACTER_DESC constant

2. Define Style Base

// Match narrative aesthetic
// Alex → "professional anime illustration, detective noir"
// Iris → "fantasy illustration, magical wonderland aesthetic"  
// Maya → "contemporary anime illustration, slice-of-life realistic"

3. Create Scenario List

// 15-20 scenarios
// Each with: id, title, scenario, attire, pose, environment, lighting, mood
// Scenarios match character's narrative role

4. Configure Output

// Directory: assets/character-reference/{character-slug}/{style-variant}
// Naming: {id}.png (e.g., "01-scenario-name.png")
// Report: generation-report.json

5. Generate Images

// For each scenario:
//   Build prompt = CHARACTER_DESC + scenario details + STYLE_BASE
//   Call Replicate API with text-only prompt
//   Handle rate limiting with exponential backoff
//   Save image and metadata

6. Validate Results

• Visual consistency check across set
• Pose variety confirmation
• Style adherence verification
• Report success/failures

Success Metrics

Consistency:

• Same character recognizable across all images
• ✅ Facial features match
• ✅ Hair color/style consistent
• ✅ Build/age appropriate

Variety:

• No duplicate poses
• ✅ Different body positions
• ✅ Varied hand gestures
• ✅ Diverse scenarios/settings

Quality:

• Professional illustration quality
• ✅ Clear details
• ✅ Proper lighting/atmosphere
• ✅ Style consistency

Efficiency:

• 100% success rate (all scenarios generated)
• Minimal retries (rate limit handling)
• Cost tracking ($0.04/image)

Common Issues & Solutions

Issue: Same pose in all images Cause: Using image_prompt without detailed pose descriptions Solution: Remove image_prompt, add explicit pose details to text prompt

Issue: Character looks different across images Cause: Vague character description Solution: Detailed, consistent character description in every prompt

Issue: Rate limit errors Cause: Too many concurrent requests Solution: 2s delay between requests + exponential backoff retry

Issue: Wrong style aesthetic Cause: Style not matching character's narrative Solution: Define STYLE_BASE appropriate to character (noir/fantasy/contemporary)

Resources

Documentation

• Flux 1.1 Pro: https://replicate.com/black-forest-labs/flux-1.1-pro
• Flux Redux (image_prompt): For composition guidance only
• Character profiles: characters/*.md

Examples

• Alex professional noir: assets/character-reference/alex/professional-noir-scenes-v2/
• Script template: scripts/generate-noir-professional-scenes.js

Cost

• Model: black-forest-labs/flux-1.1-pro
• Rate: $0.04 per image
• Typical set: 17 images = $0.68

Synaptic Connections

Related Skills

• [.github/skills/creative-writing/SKILL.md] (High, Integrates) - "Character consistency in narrative"
• [.github/skills/wonderland-worldbuilding/SKILL.md] (Medium, Enables) - "Visual representation of settings"
• [.github/skills/book-character-engine/SKILL.md] (Critical, Bidirectional) - "Character profile source"

When to Activate

• "Generate character reference images for {character}"
• "Create visual reference set"
• "Need consistent character illustrations"
• Before illustration work requiring character consistency

What This Enables

• Consistent character illustrations for chapters
• Cover art with accurate character depiction
• Marketing materials with recognizable characters
• Character design documentation