Storyloom

Storyloom

Storyloom turns public-domain fiction into a small playable experience. Upload a UTF-8 .txt book, choose a detected chapter, and explore generated scenes, clues, dialogue, and deductions as Dr. Watson.

How it works

1. The backend extracts chapters from uploaded public-domain text.
2. Long chapters are chunked, summarised, and organised into a source-grounded storyboard.
3. An LLM generates structured scene plans.
4. A reviewer pass checks source grounding, clue relevance, continuity, and unsupported deductions.
5. The backend converts validated plans into playable JSON; a browser game renders each scene on a reusable 640 × 360 2D coordinate plane.

Built with Codex and GPT-5.6

Codex and GPT-5.6 Terra supported ideation, frontend and FastAPI implementation, structured-output debugging, source-grounding rules, and demo preparation.

Judge testing

The repository includes a pre-generated Silver Blaze demo in data/storyloom.db. Judges can run the backend and static frontend locally without an API key or a new generation step; the README contains the two commands and full instructions.

Built With

• anthropic-claude
• codex
• css
• fastapi
• gpt-5.6
• groq
• html
• javascript
• pydantic
• python
• sqlite