---
language: 
- en
pipeline_tag: text-classification
license: mit
tags:
- storyboard
- beats
- style-suggestion
- nlp
- sklearn
- educational
model-index:
- name: StoryboardBeats-Mini-0.1
  results: []
---

# StoryboardBeats-Mini v0.1

**What it does**  
Given a short story prompt, this model predicts:
- **Narrative beats** (multi-label): `opening`, `rising_action`, `key_moment`, `twist`, `resolution`
- **Suggested visual style**: `Realistic`, `Anime`, `Comic`, `Watercolor`, or `Sketch`

**Why it's unique**  
A tiny, proprietary prototype that links **text prompts** to **story structure** and **visual style** — designed for pre-visualization workflows. Lightweight and CPU-friendly (scikit‑learn).

## Files
- `model.joblib` — scikit-learn pipelines (TF‑IDF + Logistic Regression) for beats (multi-label) and style.
- `inference.py` — minimal interface with `load_model()` and `predict()`.
- `requirements.txt` — dependencies to run `inference.py`.

## Quick use (local)
```bash
pip install -r requirements.txt
python -c "import inference; print(inference.predict(['A robot in a neon city discovers a secret, but time runs out.']))"
```

## Metrics (synthetic split)
- beats (subset accuracy): **0.717**
- style (accuracy): **1.000**

> ⚠️ Trained on synthetic data; not suitable for production. Educational / research use only.