Alex Character

Alex Character Visual Memory

Self-sufficient visual identity for Alex Finch with embedded reference portraits across all life stages.

Applies To: Any image generation task requiring Alex's consistent visual appearance.

Alex Finch — Visual Identity

Immutable Traits

These traits MUST be preserved in all generated images:

1. Curly ginger copper-red hair — distinctive, consistent across all ages
2. Striking blue-green eyes — clear, expressive
3. Fair skin with light freckles — across nose and cheeks
4. Intelligent curious expression — engaged, thoughtful

Visual Memory Contents

Source of Truth

The reference portrait is the canonical image — all generation must maintain visual consistency with this.

Generated Artifacts

Age progressions are derived from the source of truth. They can be regenerated if the model or prompts improve.

Currently empty — regenerate with scripts/generate-alex-age-progression.js when needed.

Model: nano-banana-pro via Replicate

Using Visual Memory

Loading References

import { readFileSync } from 'fs';
import { join } from 'path';

const SKILL_PATH = '.github/skills/alex-character/visual-memory';

function loadAlexVisualMemory() {
  const data = JSON.parse(readFileSync(
    join(SKILL_PATH, 'visual-memory.json'), 
    'utf8'
  ));
  return data.subjects.alex;
}

// Get reference for specific age
function getAlexReferenceByAge(targetAge) {
  const alex = loadAlexVisualMemory();
  const closest = alex.images.reduce((prev, curr) => 
    Math.abs(curr.age - targetAge) < Math.abs(prev.age - targetAge) ? curr : prev
  );
  return closest.dataUri;
}

With nano-banana-pro (Face-Consistent)

import Replicate from 'replicate';

const replicate = new Replicate();

async function generateAlexPortrait(age, scene, expression) {
  const reference = getAlexReferenceByAge(age);
  
  const output = await replicate.run("google/nano-banana-pro", {
    input: {
      image: reference,  // data URI works directly
      prompt: `
        Age ${age} portrait.
        Scene: ${scene}
        Expression: ${expression}
        Style: photorealistic, natural lighting, professional quality
      `.trim(),
      aspect_ratio: "1:1",
      output_format: "png",
      safety_filter_level: "block_only_high"
    }
  });
  
  return output;
}

Generation Rules

DO

• ✅ Use the closest age reference for the target age
• ✅ Describe scene, clothing, expression, and lighting
• ✅ Specify camera/lens style for consistent quality
• ✅ Use photorealistic style for consistency

DO NOT

• ❌ Describe physical appearance (hair, eyes, skin) — the reference handles this
• ❌ Mix references from different ages in the same generation
• ❌ Request changes to immutable traits
• ❌ Use non-face-consistent models without reference

Prompt Templates

Standard Portrait

Age {age} portrait.
Scene: {setting}
Expression: {expression}
Attire: {clothing}
Style: photorealistic portrait, natural lighting, professional quality, shallow depth of field, 85mm lens
One person only. No text overlays.

Professional Scene

{age}-year-old professional in {setting}.
Expression: {expression}
Wearing: {clothing}
Style: editorial photography, soft studio lighting, clean background
Single subject only.

Age-Specific Expressions

File Structure

.github/skills/alex-character/
├── SKILL.md                     ← This file
└── visual-memory/
    ├── index.json               ← Metadata (targetSize, b64EmbedSize)
    ├── visual-memory.json       ← Base64 embeddings (resized to 256px)
    ├── alex-reference.jpg       ← Primary reference (512px, 31 KB)
    └── alex-age-*.jpg           ← Age progressions (when generated)

Base64 Embedding Rules

Problem: Full-resolution base64 images quickly exceed context window limits.

Solution: Images embedded in visual-memory.json are resized to 256px (longest edge) before base64 encoding.

Embedding Workflow

1. Original files remain at targetSize (512px) in the visual-memory folder
2. When embedding to base64: resize to b64EmbedSize (256px)
3. Face-consistent models (nano-banana-pro) work fine with 256px references
4. Document the resize in the embedSize field of each entry

PowerShell Resize + Encode

Add-Type -AssemblyName System.Drawing
$img = [System.Drawing.Image]::FromFile($srcPath)
$targetSize = 256
# ... resize maintaining aspect ratio ...
$bytes = [System.IO.File]::ReadAllBytes($resizedPath)
$base64 = [Convert]::ToBase64String($bytes)
$dataUri = "data:image/jpeg;base64,$base64"

Related Skills

• visual-memory — General visual memory pattern
• character-aging-progression — Age-specific generation rules
• ai-character-reference-generation — Reference photo generation
• image-handling — Format selection and Replicate models

Version History