RolePl-AI
summary
Prototype blends rulebound procgen with local LLMs to produce coherent RPG universes: a three‑pass pipeline (outline, detail, polish) emits strict JSON schemas for worlds, cities and dialogue trees; validation rejects bad outputs; dialogue nodes carry sentiment and flag effects. Implementation uses LM Studio, an OpenAI‑compatible adapter and Hono API, runs on a local Mac Mini, serves tiny models (3 GB and 720 MB examples), enforces a two‑second timeout with fallback, and yields a playable but skeletal demo that successfully generates branching NPC dialogue.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:59AI in games past and present
- 10:40Procedural generation limits
- 13:14Generative AI as tool
- 24:15Building game worlds with AI
- 31:02Defining world constraints
- 35:42AI generation cost and efficiency
- 38:13Running AI locally with LM Studio
- 41:17Real-time generation strategies
- 44:28Game architecture and dialogue
- 47:48Live AI storytelling demo
- 50:39Closing thoughts
related talks
01h 40mOrchestrating a Role-Playing Game Universe with Multi-Container AI Agents
02h 48mCompose & Dragons: The Role-Playing Game of Agents Fed on Tiny Language Models
10mRapid Prototyping with AI
44mGlimpses of Humanity: My Game-Building AI Pair
24mPlaying D&D With AI: Building an Agentic Game Engine
47mStructured Outputs in OpenAI: Building Reliable JSON-Based AI Integrations
14mDialogLab: Authoring, Simulating, and Testing Dynamic Human-AI Group Conversations
47mGlimpses of Humanity: My Game-Building AI Pair