KurtShapes

Generative visual storytelling based on Kurt Vonnegut's insight that stories have shapes — emotional arcs that can be graphed as fortune vs. time.

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   KShapes   │     │ TinyStories │     │  Flatoons   │
│             │     │             │     │             │
│  Emotional  │ ──▶ │  Dialogue   │ ──▶ │   Visual    │
│    Arc      │     │ Generation  │     │  Rendering  │
│             │     │             │     │             │
│ curve=[...] │     │ "Oh no!"    │     │  SVG output │
└─────────────┘     └─────────────┘     └─────────────┘
     WHAT              WORDS             PICTURES
   (structure)        (text)            (visuals)

The Eight Shapes

Code	Name	Pattern
M	Man in Hole	Stable → fall → climb out better
B	Boy Meets Girl	Find good → lose it → win it back
C	Cinderella	Low → incremental rise → catastrophe → ultimate elevation
K	Kafka	Bad/neutral → worse → no redemption
H	Hamlet	Ambiguous, fortune unclear
R	Rags to Riches	Steady rise from low to high
T	Tragedy	Rise then fall, ends down
O	Old Testament	Rise to paradise → sustained fall

Current Status

295 spiffed stories across all 8 shapes, generated via LLM pipeline with panel repair and beat expression enhancement.

Shape distribution: M:11, B:25, C:13, K:70, H:54, R:19, T:52, O:51

Pipeline

Story Generation → Evaluation → Panel Repair → Dedup → Spiff Up
   (LLM+shape)     (beat/arc)   (fix panels)   (exact)  (vivid dialogue)

Project Structure

kurt/
├── CLAUDE.md              # Project instructions
├── DONE.md                # Completed work log
├── NEXT_SESSION.md        # Session handoff notes
├── README.md              # This file
├── requirements.txt       # Python dependencies
├── scripts/               # Pipeline scripts
├── results/               # Generated corpora and evaluations
├── models/                # Model artifacts
├── data/                  # Input data
├── trainingdata/          # Training data files
└── docs/                  # Documentation
    ├── KSHAPE_PLAN.md         # 7-phase implementation plan
    ├── TINYSTORIES.md         # Dialogue generation design
    ├── GOALS.md               # Project goals
    ├── PIPELINE_DESIGN.md     # Pipeline architecture
    ├── FINE_TUNING_DETAILS_STEP_BY_STEP.md
    ├── INSTRUCT_MODEL_EVALUATION.md
    ├── TINYSTORY_TRIALS.md    # Experiment trials
    ├── spiff_prompt.txt       # Enhancement prompt
    └── archive/               # Deferred specs
        ├── FLATOONS_STANDALONE.md
        ├── KSHAPE_FLATOONS.md
        ├── INTEGRATION_PLAN.md
        └── INTEGRATION_PLAN_LIGHT.md

Quick Start

pip install -r requirements.txt
export ANTHROPIC_API_KEY="your-key"

# Generate stories for a shape
python scripts/generate_story.py --shape M

# Evaluate results
python scripts/evaluate_model.py --model checkpoints/best --num-samples 100