← back
Using AI to Build an Infinite Game: Jeff Schomay
Takeaway
Cheap fine-tuning ($1-2) can replace long prompts and produce stylistically consistent generative game content at runtime.
Summary
- Built an infinite-content forest game where every scene is generated fresh by a fine-tuned OpenAI model; total fine-tune cost ~$1-2 for 50 examples.
- Fine-tuning let him shorten prompts dramatically — JSON structure became implicit in training data, reducing latency and cost.
- Used Leonardo to fine-tune a custom image model on consistent-style training data, balancing consistency (style) against variation (subject) to avoid overfitting.
- Scene definitions are JSON objects describing first-visit and return-visit content plus stat impacts.
fine-tuninggenerativegames
Original description
Making a game requires creating lots of content - tons of cheap and disposable ideas while prototyping, and a vast (or infinite?) depth of content to explore while playing. But making content is hard, expensive, and slow. AI can help if you have the right processes. In this talk I share a number of workflows and tools I used while building out both a video game and a physical card game. I show lots of examples and share what worked well and what didn't. With these tools, making the content became one of the most fun parts of making the game. Recorded & streamed live for the AI Engineer Summit 2023. See the full schedule of talks at https://ai.engineer/summit/schedule & join us at the AI Engineer World's Fair in 2024! Get your tickets today at https://ai.engineer/worlds-fair About Jeff Software engineer with a decade of experience and a passion for the intersection of games, narrative and AI.