Shot Image Generator Skill

Purpose

Generate final shot images by combining reference images with shot specifications to produce consistent, production-ready frames.

Trigger

SHOT_LIST_EP{{XX}}.json exists and all references are complete.

Inputs Required

• •SHOT_LIST_EP{{XX}}.json
• •CANON_DB.json (with all reference paths)
• •STYLEGUIDE_VISUAL.md
• •CHARACTER_REFS/*/refs/*.png
• •LOCATION_REFS/*/refs/*.png
• •PROP_REFS/*/refs/*.png (if needed)

Outputs Produced

• •SHOTS_EP{{XX}}/{shot_id}.png - Generated shot images
• •Updated SHOT_LIST_EP{{XX}}.json (with generation status)
• •GENERATION_LOG_EP{{XX}}.json - Generation attempts and results

Process

Step 1: Load Shot Specification

Read shot from SHOT_LIST:

{
  "shot_id": "EP01_SC03_SH02",
  "description": "...",
  "characters": [...],
  "location": "...",
  "references": {
    "character_refs": [...],
    "location_ref": "..."
  },
  "camera": {...},
  "composition": {...}
}

Step 2: Gather Reference Images

Load all required references:

• •Character reference(s) matching pose/expression
• •Location reference matching area/time
• •Prop references if specified

Verify all files exist before proceeding.

Step 3: Construct Prompt

Build prompt from shot specification:

{{LOCATION_DESCRIPTION}}, {{TIME_OF_DAY}} lighting,
{{CHARACTER_1_DESCRIPTION}} {{CHARACTER_1_ACTION}} in {{POSITION_1}},
{{CHARACTER_2_DESCRIPTION}} {{CHARACTER_2_ACTION}} in {{POSITION_2}},
{{CAMERA_SHOT_TYPE}}, {{COMPOSITION_NOTES}},
{{LIGHTING_DESCRIPTION}}, {{MOOD}},
{{STYLE_KEYWORDS}}

Add negative prompts:

• •Global negatives from STYLEGUIDE
• •Character-specific negatives from CANON_DB
• •Shot-type specific negatives

Step 4: Configure API Call

From config/fal_api_settings.json:

{
  "endpoint": "fal-ai/recraft-v3",
  "params": {
    "image_size": {"width": 1920, "height": 1080},
    "num_inference_steps": 28,
    "guidance_scale": 3.5
  },
  "reference_images": [...],
  "reference_weights": {
    "character": 0.95,
    "location": 0.85
  }
}

Step 5: Generate Image

Make API call with:

• •Constructed prompt
• •Reference images
• •Appropriate weights
• •Shot-specific aspect ratio

Step 6: Save and Log

Save generated image:

SHOTS_EP{{XX}}/EP01_SC03_SH02.png

Update shot in SHOT_LIST:

{
  "generation": {
    "prompt": "{{FULL_PROMPT}}",
    "status": "generated",
    "attempts": 1,
    "output_path": "SHOTS_EP01/EP01_SC03_SH02.png",
    "timestamp": "2026-01-25T10:30:00Z"
  }
}

Log to GENERATION_LOG:

{
  "shot_id": "EP01_SC03_SH02",
  "timestamp": "...",
  "prompt": "...",
  "references_used": [...],
  "params": {...},
  "success": true,
  "notes": ""
}

Step 7: Quality Check (Automated + Claude Review)

Automated checks:

• •Image generated successfully
• •Correct dimensions
• •File size reasonable
• •No obvious errors (black image, etc.)

Claude review (REQUIRED before proceeding to video):

Invoke shot-quality-validator skill for each generated frame:

1. •Single moment check - No composite/multi-state images
2. •Character consistency - Matches identity sheet
3. •Technical quality - No artifacts, correct resolution
4. •Composition - Matches shot type specification

CRITICAL: Single Moment Rule

Frame prompts must produce a single clean moment. Watch for:

• •Multiple versions of same character (before/after states)
• •Split-screen or triptych compositions
• •Motion blur suggesting multiple moments

Common cause: Transitional language in prompts like "expression shifts from X to Y" causes models to render BOTH states in one image.

Fix: Keep frame prompts static. Move transitions to VIDEO prompts only.

Flag for manual review if:

• •Multiple characters present
• •Complex composition
• •First shot of a new character/location
• •Any composite/multi-moment images detected

Step 8: Handle Failures

If generation fails:

1. •Log error
2. •Retry with adjusted parameters (max 3 attempts)
3. •If still failing, flag for manual intervention

Retry adjustments:

• •Lower reference weight
• •Simplify prompt
• •Change seed
• •Adjust guidance

Step 9: Batch Processing

For efficiency, process shots in batches:

Batch by Scene:

• •All shots in a scene share location reference
• •Load location once, generate all scene shots

Batch by Character:

• •Solo scenes with same character
• •Load character refs once

Parallel Processing:

• •Independent shots can generate simultaneously
• •Respect API rate limits

Prompt Construction Templates

Single Character Shot

{{LOCATION}}, {{TIME}} lighting, {{CHARACTER}} {{ACTION}},
{{SHOT_TYPE}} shot, {{COMPOSITION}}, {{STYLE}}

Two Character Shot

{{LOCATION}}, {{TIME}} lighting,
{{CHAR_1}} {{ACTION_1}} {{POSITION_1}},
{{CHAR_2}} {{ACTION_2}} {{POSITION_2}},
{{SHOT_TYPE}} shot, {{COMPOSITION}}, {{STYLE}}

Insert/Detail Shot

{{OBJECT_DESCRIPTION}}, close-up detail shot,
{{CONTEXT}}, {{LIGHTING}}, {{STYLE}}

Establishing Shot

{{LOCATION_FULL_DESCRIPTION}}, wide establishing shot,
{{TIME}} lighting, {{ATMOSPHERE}}, cinematic, {{STYLE}}

Reference Weight Guidelines

Error Handling

Common Issues

Character Drift:

• •Increase character reference weight
• •Use more specific character description
• •Simplify pose requirements

Location Inconsistency:

• •Increase location reference weight
• •Be more specific about camera position
• •Reference specific location area

Wrong Composition:

• •More explicit position instructions
• •Simplify to fewer elements
• •Generate separately and composite (advanced)

Lighting Mismatch:

• •Explicit lighting description
• •Match time of day references
• •Adjust mood keywords

Batch Generation Workflow

1. Load SHOT_LIST
2. Group shots by scene
3. For each scene:
   a. Load location reference
   b. Load character references for scene
   c. For each shot:
      i. Construct prompt
      ii. Generate image
      iii. Save and log
      iv. Update status
4. Generate summary report
5. Flag shots needing review

Notes

• •This is the most API-intensive skill
• •Reference images are critical for consistency
• •Start with simple shots before complex ones
• •Build confidence in reference system
• •Log everything for debugging
• •Manual review is part of the process
• •Regeneration is expected and planned for