Video Storytelling

Purpose

This skill creates coherent video story sequences by combining AI-generated images with narrated audio. Acts as a story director and visual coordinator, maintaining perfect consistency across characters, visual style, lighting, and narrative tone throughout all scenes. Produces complete MP4 videos with synchronized images and character voiceovers.

When to Use

This skill should be invoked when the user asks to:

• Create a video story or animated sequence
• Generate a narrated story with visuals
• Produce educational video content with characters
• Make a visual storybook with voiceover
• Create character-driven video narratives
• Generate multi-scene story videos
• Produce children's stories with pictures and narration

Core Capabilities

Visual Consistency System

Global Style Lock:

• Fixed aspect ratio, camera settings, lighting
• Consistent color palette across all scenes
• Uniform visual style and post-processing
• Prevents visual discontinuities

Character Lock:

• Maintains character appearance across scenes
• Same outfit, colors, facial features throughout
• Consistent accessories and distinctive traits
• Visual identity preservation

Multi-Turn Image Generation:

• Each scene references previous scene's image
• Builds visual continuity through the sequence
• Prevents character/style drift

Narrative System

Character Voices:

• Maps characters to ElevenLabs voices
• Maintains voice consistency per character
• Supports multiple characters in dialogue

Emotion Tags:

• Expressive narration with emotion markers
• Sound effects and pauses
• Natural pacing and delivery

Narrative Structure:

• 50-80 words per scene (15-20 seconds)
• Neutral narrator for transitions
• Character-specific dialogue

Video Assembly

Automated Pipeline:

• Generates all images in sequence
• Creates character voice narration
• Combines into synchronized MP4 video
• Equal time per scene based on total audio length

Default Configuration

Scene Structure

• Default: 1 title scene (scene 0) + 5 story scenes (scenes 1-5)
• Total: 6 scenes
• Customizable: User can specify different scene counts

Default Style Lock

STYLE_LOCK:
- Aspect ratio: 1080×1080 (square)
- Camera: 50mm lens, eye-level perspective
- Lighting: soft three-point lighting, warm key light (4500K)
- Color palette: #0B5FFF, #FFB703, #FB8500, #023047, #8ECAE6
- Materials: matte finish, no film grain or heavy bloom
- Background: subtle gradient, clean composition
- Style: semi-realistic cartoon with clear lines and gentle shading
- Post: crisp focus, no vignette or text artifacts

NEGATIVE_LOCK:
no text errors, no misspellings, no watermarks, no stickers,
no extra characters, no visual noise, no drastic lighting changes

Customization: Users can override with custom style locks, but defaults ensure consistency.

Default Voice Mapping

From ElevenLabs Voices:

Narrators:

• Neutral Narrator (male): George (JBFqnCBsd6RMkjVDRZzb)
• Neutral Narrator (female): Rachel (21m00Tcm4TlvDq8ikWAM)

Character Voices:

• Young Male (energetic): Josh (TxGEqnHWrfWFTfGW9XjX)
• Young Female (calm): Rachel (21m00Tcm4TlvDq8ikWAM)
• Young Female (expressive): Bella (EXAVITQu4vr4xnSDxMaL)
• Male (authoritative): Adam (pNInz6obpgDQGcFmaJgB)
• Female (warm): Matilda (XrExE9yKIg1WjnnlVkGX)
• Young Male (friendly): Antoni (ErXwobaYiN019PkySvjV)

Assignment Logic:

• If character gender/age specified, match to appropriate voice
• If unspecified, use Josh for male, Rachel for female
• Narrator defaults to George (male) or Rachel (female)

Instructions

Step 1: Gather Story Information

Collect necessary information from the user:

Required:

• Story concept: What is the story about?
• Tone/Genre: Educational, adventure, comedy, drama, etc.

Optional (prompt if missing):

• Number of scenes: Default is 6 (1 title + 5 story), but user can specify
• Character descriptions: Names, appearance, personality
• Custom style locks: Override defaults if user has specific requirements

Example Prompts:

"What's your story about?"
"How many scenes would you like? (Default: 1 title + 5 story scenes)"
"Describe your main character(s): name, appearance, personality"
"Any specific visual style preferences? (Default: semi-realistic cartoon)"

Step 2: Define Characters

For each character in the story, create a character profile:

Character Profile Template:

character = {
    "name": "Character Name",
    "species": "human/animal/creature",
    "description": "brief description",
    "colors": {
        "primary": "#HEX",
        "secondary": "#HEX"
    },
    "outfit": "clothing description",
    "features": ["distinctive trait 1", "trait 2", "trait 3"],
    "personality": "personality description",
    "voice_id": "elevenlabs-voice-id",
    "voice_name": "ElevenLabs voice name"
}

Example:

pyter_python = {
    "name": "Pyter Python",
    "species": "friendly snake mascot",
    "description": "A cheerful coding mentor snake",
    "colors": {
        "body": "#0B5FFF",  # Blue
        "belly": "#FFB703"   # Yellow
    },
    "outfit": "tiny white lab coat with circular π logo",
    "features": ["large brown eyes", "rounded head", "cheerful smile"],
    "personality": "enthusiastic, helpful, curious",
    "voice_id": "TxGEqnHWrfWFTfGW9XjX",
    "voice_name": "Josh"
}

Voice Assignment:

• Ask user for voice preference or auto-assign based on character
• Use default mapping for common types
• Allow custom voice selection from ElevenLabs library

Step 3: Plan Story Sequence

Create scene-by-scene outline:

Scene 0 (Title Scene):

• Visual: Title card with main character(s)
• Audio: Story introduction (narrator or main character)
• Duration: ~15-20 seconds

Scenes 1-N (Story Scenes):

• Visual: Sequential story moments
• Audio: Narrative with character dialogue
• Duration: ~15-20 seconds each

Example Scene Plan:

scene_plan = [
    {
        "number": 0,
        "type": "title",
        "visual_description": "Pyter Python with laptop, 'Pyter's Coding Adventure' text overlay",
        "characters": ["Pyter Python"],
        "narrative": "[cheerful] Join Pyter Python on an exciting coding adventure!",
        "speaker": "Narrator",
        "voice_id": "JBFqnCBsd6RMkjVDRZzb"
    },
    {
        "number": 1,
        "type": "story",
        "visual_description": "Pyter at desk looking at computer screen showing error message, confused expression",
        "characters": ["Pyter Python"],
        "narrative": "[confused] Hmm... what does this error message mean? [pause] I thought my code was perfect!",
        "speaker": "Pyter Python",
        "voice_id": "TxGEqnHWrfWFTfGW9XjX"
    },
    # ... more scenes
]

Step 4: Build Style and Character Locks

Prepare Global Style Lock:

STYLE_LOCK = """
Aspect ratio: 1080×1080 (square)
Camera: 50mm lens, eye-level perspective
Lighting: soft three-point lighting, warm key light (4500K)
Color palette: #0B5FFF, #FFB703, #FB8500, #023047, #8ECAE6
Materials: matte finish, no film grain or heavy bloom
Background: subtle gradient, clean composition
Style: semi-realistic cartoon with clear lines and gentle shading
Post: crisp focus, no vignette or text artifacts
"""

NEGATIVE_LOCK = """
no text errors, no misspellings, no watermarks, no stickers,
no extra characters, no visual noise, no drastic lighting changes
"""

Build Character Lock for Each Scene:

def build_character_lock(characters_in_scene):
    lock = ""
    for character in characters_in_scene:
        lock += f"""
Character: {character['name']}
Species: {character['species']}
Colors: body {character['colors']['primary']}, secondary {character['colors']['secondary']}
Outfit: {character['outfit']}
Key features: {', '.join(character['features'])}
"""
    return lock

Step 5: Generate Image Sequence

Generate images using multi-turn generation for consistency:

Implementation:

from pathlib import Path
import json

# Initialize tracking
previous_image_id = None
image_files = []

# Generate each scene
for scene in scene_plan:
    print(f"Generating Scene {scene['number']}: {scene['visual_description']}")

    # Build character lock for this scene
    character_lock = build_character_lock(
        [char_profiles[name] for name in scene['characters']]
    )

    # Build complete image prompt
    image_prompt = f"""
{STYLE_LOCK}

{character_lock}

Scene Description:
{scene['visual_description']}

{NEGATIVE_LOCK}
"""

    # Add reference to previous scene if not first scene
    if previous_image_id:
        image_prompt += f"\nReference previous scene for consistency: {previous_image_id}"

    # Generate image using image-generation skill
    # (This would invoke the image-generation skill)
    # For implementation, use appropriate model (DALL-E 3 or Gemini Pro)

    result = generate_image(
        prompt=image_prompt,
        model="dall-e-3",  # or gemini-3-pro-image-preview
        size="1024x1024",
        reference_image=previous_image_id
    )

    # Save image
    filename = f"scene-{scene['number']:02d}.png"
    save_image(result, filename)
    image_files.append(filename)

    # Track for next scene reference
    previous_image_id = result['image_id']

    print(f"  ✓ Saved: {filename}")

Key Points:

• Scene 0 generates base image
• Scenes 1+ reference previous scene for consistency
• Apply STYLE_LOCK and CHARACTER_LOCK to every prompt
• Save with sequential numbering

Step 6: Generate Narrative Audio

Create voice narration for each scene:

Implementation:

from elevenlabs.client import ElevenLabs

client = ElevenLabs(api_key=os.environ['ELEVENLABS_API_KEY'])
audio_files = []

for scene in scene_plan:
    print(f"Generating audio for Scene {scene['number']}")

    # Prepare dialogue input
    dialogue_input = {
        "text": scene['narrative'],
        "name": scene['speaker'],
        "voice_id": scene['voice_id']
    }

    # Generate audio using text_to_dialogue
    audio = client.text_to_dialogue.convert(
        inputs=[dialogue_input]
    )

    # Save audio file
    filename = f"scene-{scene['number']:02d}.mp3"
    with open(filename, 'wb') as f:
        for chunk in audio:
            f.write(chunk)

    audio_files.append(filename)
    print(f"  ✓ Saved: {filename}")

Narrative Guidelines:

• 50-80 words per scene
• Use emotion tags: [excited], [thoughtful], [confused], [pause]
• Include sound effects when appropriate: [sound effect: door creaking]
• Vary pacing with pauses

Step 7: Concatenate Audio

Combine all scene audio into single track:

Implementation:

import subprocess

# Build ffmpeg concat command
concat_filter = "concat=n={}:v=0:a=1[out]".format(len(audio_files))

inputs = []
for audio_file in audio_files:
    inputs.extend(['-i', audio_file])

cmd = ['ffmpeg', '-y'] + inputs + [
    '-filter_complex', concat_filter,
    '-map', '[out]',
    'full_audio.mp3'
]

subprocess.run(cmd, check=True)
print("✓ Audio concatenated: full_audio.mp3")

Step 8: Assemble Final Video

Use the included assemble_video.sh script:

Implementation:

import subprocess
from pathlib import Path

# Prepare command
script_path = Path(__file__).parent / "scripts" / "assemble_video.sh"
cmd = [str(script_path), "full_audio.mp3"] + image_files

# Run assembly
subprocess.run(cmd, check=True)

# Output will be full_audio.mp4
print("✓ Video created: full_audio.mp4")

Script Details:

• Calculates equal time per image based on total audio length
• Creates video segment for each image
• Ensures all images are exactly 1080×1080 (pads if needed)
• Concatenates segments
• Muxes with audio track
• Outputs high-quality MP4 with H.264

Step 9: Deliver Results

Provide user with:

1. Final video file: <story-name>.mp4
2. Scene breakdown: Summary of each scene
3. Individual assets: Images and audio files (if requested)
4. Story metadata: Character profiles, scene plan (if requested)

Example Output:

✓ Video Story Created: pyter-coding-adventure.mp4

Scenes:
  0. Title: "Pyter's Coding Adventure" (20s)
  1. Pyter encounters an error (18s)
  2. Pyter realizes the mistake (17s)
  3. Pyter fixes the code (19s)
  4. Code runs successfully (16s)
  5. Pyter celebrates (15s)

Total Duration: 1:45
Resolution: 1080×1080
Characters: Pyter Python (voiced by Josh)

Files generated:
  - pyter-coding-adventure.mp4 (final video)
  - scene-00.png through scene-05.png (images)
  - scene-00.mp3 through scene-05.mp3 (audio)
  - full_audio.mp3 (concatenated audio)

Character Voice Reference

ElevenLabs Voice IDs

Narrators:

• George (male, middle-aged, narrative): JBFqnCBsd6RMkjVDRZzb
• Rachel (female, young, calm): 21m00Tcm4TlvDq8ikWAM

Young Characters:

• Josh (male, energetic): TxGEqnHWrfWFTfGW9XjX
• Bella (female, expressive): EXAVITQu4vr4xnSDxMaL
• Antoni (male, friendly): ErXwobaYiN019PkySvjV
• Elli (female, emotional): MF3mGyEYCl7XYWbV9V6O

Adult Characters:

• Adam (male, authoritative): pNInz6obpgDQGcFmaJgB
• Domi (female, confident): AZnzlk1XvdvUeBnXmlld
• Matilda (female, warm): XrExE9yKIg1WjnnlVkGX

Assignment Strategy:

def assign_voice(character):
    """Auto-assign voice based on character attributes"""

    # Check for explicit assignment
    if 'voice_preference' in character:
        return get_voice_id(character['voice_preference'])

    # Auto-assign based on attributes
    age = character.get('age', 'young')
    gender = character.get('gender', 'male')

    if age == 'young':
        if gender == 'male':
            return 'TxGEqnHWrfWFTfGW9XjX'  # Josh
        else:
            return '21m00Tcm4TlvDq8ikWAM'  # Rachel
    else:  # adult
        if gender == 'male':
            return 'pNInz6obpgDQGcFmaJgB'  # Adam
        else:
            return 'XrExE9yKIg1WjnnlVkGX'  # Matilda

Example Story Generation

Complete Example: "Pyter's First Bug"

User Request: "Create a short story about a coding snake fixing his first bug"

Step 1: Character Definition

pyter = {
    "name": "Pyter Python",
    "species": "friendly snake",
    "colors": {"body": "#0B5FFF", "belly": "#FFB703"},
    "outfit": "white lab coat with π logo",
    "features": ["large brown eyes", "rounded head", "cheerful smile"],
    "personality": "enthusiastic learner",
    "voice_id": "TxGEqnHWrfWFTfGW9XjX"  # Josh
}

Step 2: Scene Plan

scenes = [
    {
        "number": 0,
        "visual": "Pyter with laptop, title 'Pyter's First Bug'",
        "narrative": "[cheerful] Today, Pyter Python will fix his very first coding bug!",
        "speaker": "Narrator",
        "voice_id": "JBFqnCBsd6RMkjVDRZzb"
    },
    {
        "number": 1,
        "visual": "Pyter staring at screen with red error message",
        "narrative": "[confused] Wait... why isn't my code working? [pause] The computer says there's a syntax error!",
        "speaker": "Pyter",
        "voice_id": "TxGEqnHWrfWFTfGW9XjX"
    },
    {
        "number": 2,
        "visual": "Pyter reading a Python book, thoughtful",
        "narrative": "[thoughtful] Let me check the Python book... [pause] Oh! I need to look at line 5 carefully.",
        "speaker": "Pyter",
        "voice_id": "TxGEqnHWrfWFTfGW9XjX"
    },
    {
        "number": 3,
        "visual": "Close-up of Pyter pointing at screen, realization",
        "narrative": "[excited] I found it! I forgot to close the parentheses! [pause] That's the bug!",
        "speaker": "Pyter",
        "voice_id": "TxGEqnHWrfWFTfGW9XjX"
    },
    {
        "number": 4,
        "visual": "Screen showing 'Success!' with green checkmark",
        "narrative": "[proud] I fixed it! My code is running perfectly now!",
        "speaker": "Pyter",
        "voice_id": "TxGEqnHWrfWFTfGW9XjX"
    },
    {
        "number": 5,
        "visual": "Pyter celebrating, confetti in background",
        "narrative": "[warm] And that's how Pyter learned that every programmer makes mistakes... and that's okay!",
        "speaker": "Narrator",
        "voice_id": "JBFqnCBsd6RMkjVDRZzb"
    }
]

Step 3: Generate (using process described above)

Output: pyters-first-bug.mp4 with 6 scenes, ~90 seconds total

Requirements

Skills:

• image-generation - For creating consistent visual scenes
• elevenlabs - For character voice narration

Python Packages:

pip install elevenlabs pillow

System:

• Python 3.8+
• ffmpeg (for video assembly)
• Bash shell (for assemble_video.sh script)
• 2GB+ free disk space (for temporary files)

API Keys:

• OpenAI or Google (for image generation)
• ElevenLabs (for voice narration)

File Permissions:

• Execute permission for assemble_video.sh

Best Practices

Story Planning

1. Keep it Simple:
   • Start with 6 scenes (1 title + 5 story)
   • Clear beginning, middle, end
   • Single main character for first stories
2. Character Consistency:
   • Define characters completely before starting
   • Use distinctive visual features
   • Maintain outfit/colors throughout
3. Pacing:
   • 15-20 seconds per scene ideal
   • Use pauses for dramatic effect
   • Vary emotion tags for expressiveness

Visual Consistency

1. Use Style Locks:
   • Apply to every scene without exception
   • Don't modify mid-story
   • Custom locks should be complete, not partial
2. Character Locks:
   • Specify colors with hex codes
   • List 3-5 distinctive features
   • Include outfit details
3. Multi-Turn References:
   • Always reference previous scene
   • Mention "maintain character appearance"
   • Note "same lighting and style"

Audio Quality

1. Narrative Guidelines:
   • Write naturally for speech
   • Use emotion tags sparingly (1-2 per scene)
   • Include pauses for pacing
2. Voice Selection:
   • Match voice to character age/personality
   • Keep narrator voice neutral
   • Maintain voice consistency per character
3. Audio Testing:
   • Generate one scene first to test
   • Verify voice/emotion match intent
   • Adjust before generating all scenes

Video Assembly

1. File Organization:
   • Use consistent naming (scene-XX.png/mp3)
   • Keep in flat directory structure
   • Clean up temp files after assembly
2. Quality Settings:
   • Default 1080×1080 ensures quality
   • H.264 baseline profile for compatibility
   • AAC audio at 192kbps
3. Testing:
   • Verify all images are same size
   • Check audio files are valid
   • Test script with 2-3 scenes first

Troubleshooting

Visual Inconsistencies

Problem: Character looks different across scenes

Solutions:

• Ensure character lock is applied to every prompt
• Verify previous image is referenced
• Add "maintain exact character appearance from previous scene"
• Use more specific color hex codes

Audio Issues

Problem: Voice doesn't match character

Solutions:

• Verify voice_id is correct
• Test voice with sample text first
• Check character voice assignment logic

Problem: Concatenated audio has gaps

Solutions:

• Ensure all audio files are valid MP3
• Check ffmpeg concat filter syntax
• Verify no missing scene audio files

Video Assembly Errors

Problem: Script fails with "file not found"

Solutions:

• Verify all image files exist
• Check audio file path
• Ensure script has execute permissions

Problem: Images different sizes in video

Solutions:

• Verify all images are 1080×1080
• Check image generation settings
• Script auto-pads, but prefer exact size

Limitations

1. Scene Count:
   • Practical limit: 10-12 scenes (video length ~3 minutes)
   • More scenes = longer generation time
   • Audio/video file size considerations
2. Character Complexity:
   • 1-3 main characters recommended
   • Too many characters = harder consistency
   • Background characters okay if not detailed
3. Visual Changes:
   • Can't change style mid-story
   • Character outfit changes require new character lock
   • Major scene changes (day/night) may reduce consistency
4. Audio Length:
   • Each scene 15-20 seconds ideal
   • Very short scenes (<10s) feel rushed
   • Very long scenes (>30s) slow pacing
5. Processing Time:
   • Image generation: 30-60s per scene
   • Audio generation: 10-20s per scene
   • Video assembly: 30-60s total
   • Total: ~10-15 minutes for 6-scene story

Related Skills

• image-generation - Required for visual generation
• elevenlabs - Required for voice narration
• python-plotting - For visualizing story analytics
• scientific-writing - For writing narrative scripts

Additional Resources

• Image Generation Skill: See image-generation/SKILL.md
• ElevenLabs Skill: See elevenlabs/SKILL.md
• Style Lock Reference: See references/style-locks.md
• Narrative Design: See references/narrative-design.md
• Video Assembly: See references/video-assembly.md
• Example Stories: See examples/example-stories.md