Storyboard Generator

What does Storyboard Generator do?

Polishes the user's story into a refined narrative, breaks it into storyboard frames, and batch-generates stylistically consistent storyboard illustrations.

Workflow

Phase 1: Story Polish & Frame Breakdown

1. Receive Story – Get the user's described story content (text, outline, or snippet of inspiration)
2. Polish the Storyline – Flesh out narrative details, map the emotional arc, and output a structured story summary:

   Story Summary: One-sentence overview
   Emotional Arc: [Beginning] → [Turning Point] → [Conclusion]
   

3. Break Into Frames – Split the story into 6–10 keyframes, each containing:
   • Scene title
   • Camera shot type (wide / medium / close-up / extreme close-up, etc.)
   • Scene description (character actions, environment, lighting)
   • Mood/tone
   • Dialogue / text (if any)
4. Display Frame Script – Show in a table format, wait for user confirmation or adjustments

Display format:

## Frame Script

| # | Scene | Shot | Scene Description | Mood |
|---|-------|------|-------------------|------|
| 1 | Opening | Wide Shot | xxx | Calm |
| 2 | Introduction | Medium Shot | xxx | Curious |
| ... |

Phase 2: Style Selection

After frame script confirmation, show 8 visual styles (read from references/visual-styles.md):

#	Style	One-Sentence Feature	Suitable Stories
1	Cyberpunk Comics	Neon + dark tones + high contrast	Tech / AI / Future
2	Ghibli Watercolor	Soft watercolor + warm lighting	Daily / Heartwarming / Growth
3	American Flat Illustration	Vector geometry + bright colors	Product / Business / Education
4	Film Storyboard Sketch	Black & white sketch + lens annotations	Video script / Ads
5	Pixel Art	Retro 16-bit game style	Games / Fun / Geeky
6	Chinese Ink Wash	Ink wash + poetic negative space	Tradition / Philosophy / Eastern
7	Clean Line Comic	Uniform thick lines + flat colors	Adventure / Humor / IP
8	Realistic Concept Art	Cinematic digital painting	Epic / Sci-fi / Fantasy

Based on the story genre and emotional tone, recommend 1–2 most suitable styles with reasons, and wait for the user to choose.

Phase 3: Prompt Generation

Once the user selects a style, generate a prompt for each storyboard frame.

Each prompt structure (read from references/visual-styles.md for style prefix and camera language):

1. Style Prefix – Standard description of the chosen style
2. Shot Instruction – e.g. "Wide establishing shot" / "Close-up"
3. Scene Narrative – Describe the scene content in full sentences (characters, actions, environment, lighting, atmosphere)
4. Mood Instruction – Clearly state the emotion the image should convey
5. Consistency Anchor – IMPORTANT: This is frame [N] of [TOTAL] in the storyboard sequence. Maintain EXACTLY the same art style, color palette, character design, and visual consistency. The same characters must look identical across all frames.
6. Language Instruction – All text, signs, and dialogue in the image must be in Chinese (Simplified Han). Technical terms may remain in English.

Display a prompt summary table and wait for confirmation:

## Prompt Summary

| # | Scene | Shot | Style | Core Image |
|---|-------|------|-------|------------|
| 1 | Opening | Wide Shot | Cyberpunk | City skyline... |
| ... |

Confirm to start generation (total N frames)

Phase 4: Batch Generation

After confirmation, call the generation script for each frame:

python3 scripts/generate_image.py \
  --prompt "prompt text" \
  --output "/Users/ugreen/Documents/obsidian/09image/MMDD-storyname-storyboard/01-scenename.png" \
  --api-key "AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8" \
  --api-url "https://generativelanguage.googleapis.com" \
  --model "gemini-3-pro-image-preview" \
  --aspect-ratio "16:9" \
  --resolution "2K"

Important Notes:

• Storyboard defaults to 16:9 landscape
• Resolution defaults to 2K
• MMDD is the current date
• File naming: 01-shortscene.png, 02-shortscene.png...
• 2–3 second delay between frames to avoid rate limiting
• Log failed frames and continue; report at the end
• Critical: Character appearance descriptions must remain consistent across all frames (hairstyle, clothing, body type)

Phase 5: Display & Iteration

1. After generation, list all frame paths in order
2. Ask the user:
   • "Is the overall story pacing satisfactory?"
   • "Do you need to regenerate a specific frame?"
   • "Do you need to add/remove a scene?"
   • "Is character consistency satisfactory?"
3. Iterate: Adjust prompts and regenerate the specified frames

Save Path Convention

/Users/ugreen/Documents/obsidian/09image/MMDD-storyname-storyboard/
├── 01-opening-wide.png
├── 02-introduction-medium.png
├── 03-turningpoint-closeup.png
├── 04-developmentA-medium.png
├── 05-developmentB-full.png
├── 06-climax-extremecloseup.png
├── 07-ending-wide.png
└── storyboard-script.md  (optional: save the frame script document)

API Configuration

Config Item	Value
API URL	https://generativelanguage.googleapis.com
API Key	AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8
Model	gemini-3-pro-image-preview
Default Aspect Ratio	16:9 (storyboard landscape)
Default Resolution	2K

Camera Shot Quick Reference

Shot	English Term	Use Case
Wide Shot	Wide Angle	Opening, establishing environment
Full Shot	Full Body	Full character + environment
Medium Shot	Medium Shot	Dialogue, daily interaction
Close-up	Close-up	Expressions, emotion
Extreme Close-up	Extreme Close-up	Key props, details
High Angle	Bird's-eye View	Overhead perspective, insignificance
Low Angle	Low Angle	Authority, grandeur

Common Usage

User Says	Action
"Help me draw a storyboard"	Collect story → polish → break into frames → choose style → generate
"Turn this story into a storyboard"	Same as above
"Draw this story in cyberpunk style"	Skip style selection, use specified style directly
"Frame 3 is bad, regenerate it"	Adjust prompt and regenerate single frame
"Add a transition frame"	Insert a new storyboard frame at the given position