Character Reference Generator Skill

Purpose

Generate comprehensive character reference image packs that enable consistent character reproduction across all shots.

Trigger

CANON_DB.json and STYLEGUIDE_VISUAL.md exist.

Inputs Required

• •CANON_DB.json (character visual data)
• •STYLEGUIDE_VISUAL.md
• •CHARACTER_SHEETS/*.md

Outputs Produced

• •CHARACTER_REFS/{NAME}/VISUAL_SPEC.md - Character visual specification
• •CHARACTER_REFS/{NAME}/refs/*.png - Reference images
• •Updated CANON_DB.json (reference paths added)

Reference Types Required

For each main character, generate:

1. Turnaround Sheet

• •Front view (neutral pose)
• •3/4 view (neutral pose)
• •Profile view (neutral pose)
• •Back view (neutral pose)

Purpose: Establish consistent proportions and features from all angles.

2. Expression Pack

• •Neutral
• •Happy/Pleased
• •Sad/Disappointed
• •Angry/Frustrated
• •Surprised/Shocked
• •Determined
• •Vulnerable

Purpose: Establish how features change with emotion while remaining recognizable.

3. Outfit Variants

• •Primary/Signature outfit
• •2-3 alternate outfits (based on script needs)

Purpose: Enable wardrobe changes while maintaining character identity.

4. Action Poses

• •3-5 key poses from script scenarios
• •e.g., running, fighting, working, relaxed

Purpose: Reference for dynamic shots.

5. Detail Close-ups

• •Face/head detail
• •Hands (if signature gestures)
• •Signature props with character

Purpose: High-detail reference for close-up shots.

Process

Step 1: Extract Character Data

From CANON_DB.json, retrieve:

{
  "description": "...",
  "signature_outfit": "...",
  "color_palette": [...],
  "signature_prop": "...",
  "negative_prompts": [...]
}

Step 2: Create Visual Specification

Generate CHARACTER_REFS/{NAME}/VISUAL_SPEC.md:

# Visual Specification: {{CHARACTER_NAME}}

## Physical Description
{{FROM_CANON_DB}}

## Signature Look
- **Outfit**: {{OUTFIT}}
- **Colors**: {{PALETTE}}
- **Props**: {{PROPS}}

## Visual Anchors (LOCKED)
- {{FEATURE_1}}
- {{FEATURE_2}}
- {{FEATURE_3}}

## Allowed Variations
- {{VARIATION_1}}
- {{VARIATION_2}}

## Negative Prompts
- {{NEGATIVE_1}}
- {{NEGATIVE_2}}

## Reference Status
- [ ] Turnaround
- [ ] Expressions
- [ ] Outfits
- [ ] Poses
- [ ] Details

Step 3: Generate Turnaround (SEED IMAGE)

Load prompt template:

prompts/turnaround.txt

Construct prompt:

1. •Insert character description
2. •Insert signature outfit
3. •Add style keywords from STYLEGUIDE_VISUAL.md
4. •Add negative prompts

API Call (fal.ai/recraft-v3):

{
  "prompt": "{{CONSTRUCTED_PROMPT}}",
  "negative_prompt": "{{NEGATIVE_PROMPTS}}",
  "image_size": {"width": 1024, "height": 1024},
  "num_inference_steps": 28,
  "guidance_scale": 3.5
}

Save result as refs/{name}_turnaround.png

CRITICAL: This becomes the SEED reference for all subsequent images.

Step 4: Generate Expression Pack

Use turnaround as reference input.

For each expression:

1. •Load prompts/expressions.txt
2. •Modify for specific expression
3. •Include turnaround as reference image
4. •Set reference_weight: 0.95

Save as refs/{name}_expr_{emotion}.png

Step 5: Generate Outfit Variants

Use turnaround as reference input.

For each outfit:

1. •Load prompts/outfits.txt
2. •Modify for specific outfit description
3. •Include turnaround as reference image
4. •Set reference_weight: 0.90 (allow outfit variation)

Save as refs/{name}_outfit_{variant}.png

Step 6: Generate Action Poses

Use turnaround as reference input.

For each pose:

1. •Load prompts/action_poses.txt
2. •Describe specific pose
3. •Include turnaround as reference image
4. •Set reference_weight: 0.85 (allow pose variation)

Save as refs/{name}_pose_{action}.png

Step 7: Generate Details

Use turnaround and best expression as references.

1. •Face close-up
2. •Hands (if needed)
3. •Signature prop interaction

Save as refs/{name}_detail_{type}.png

Step 8: Update CANON_DB

Add reference paths to character entry:

"reference_images": {
  "turnaround": "CHARACTER_REFS/ALICE_CHEN/refs/alice_turnaround.png",
  "front_neutral": "CHARACTER_REFS/ALICE_CHEN/refs/alice_front_neutral.png",
  ...
}

Step 9: Quality Check

For each generated image:

• • Features match turnaround
• • Expression is readable
• • Outfit is accurate
• • Pose is natural
• • Style matches STYLEGUIDE

Flag failures for regeneration.

Prompt Templates

Located in prompts/:

turnaround.txt

See template file for full prompt structure.

expressions.txt

See template file for expression-specific prompts.

outfits.txt

See template file for outfit variant prompts.

action_poses.txt

See template file for dynamic pose prompts.

API Configuration

Located in config/fal_api_settings.json:

{
  "endpoint": "fal-ai/recraft-v3",
  "default_params": {
    "image_size": {"width": 1024, "height": 1024},
    "num_inference_steps": 28,
    "guidance_scale": 3.5,
    "safety_tolerance": "2",
    "output_format": "png"
  },
  "reference_mode_params": {
    "style_reference_weight": 0.85,
    "character_reference_weight": 0.95
  }
}

Notes

• •Turnaround is the SEED - all other images reference it
• •Higher reference weights = more consistency, less variation
• •Lower reference weights = allow for pose/outfit changes
• •Always review generated images before proceeding
• •Regenerate any images that break visual anchors
• •Update CANON_DB after successful generation