Creating an Infinite Exploration Game with AI-Generated Content: A Unique Generative AI Project

Exploring the infinite possibilities of generative AI, I created a unique game where every scene is dynamically generated, offering endless exploration and discovery.

  • 1. Jeff created a game with 100% AI-generated content.
  • 2. The game is about wandering in a forest, going from scene to scene, and trying to find your home before running out of courage.
  • 3. There are 16 scenes in a 4x4 grid, and every game is unique as scenes are generated fresh each time.
  • 4. Jeff's favorite part was generating each scene and seeing the AI-created content.
  • 5. He used OpenAI's completion endpoint for generating scenes with prompt engineering.
  • 6. To make the generation more reliable, he fine-tuned a model using OpenAI's fine-tuning endpoint and 50-100 examples.
  • 7. The fine-tuned model required a shorter and simpler prompt, focusing on general scene description rather than specific JSON format.
  • 8. For generating images, Jeff used Leonardo, which allows custom image models for stylistic consistency.
  • 9. He experimented with various parameters and prompts to create suitable images matching the scenes.
  • 10. Fine-tuning an image model required consistent images with the same parts but varied content; this was challenging due to different scene elements.
  • 11. Jeff trained multiple models with different parameters and sets of images before finding a suitable one.
  • 12. Generated images show similar features and style while maintaining variety in scene details.
  • 13. An AI pipeline was created using OpenAI's custom model for scenes and Leonardo for images.
  • 14. The pipeline includes JSON validation, image generation, and putting it all together for the game.
  • 15. Caching was added to handle the time it takes to generate new scenes, ensuring there's always a ready-to-use scene available.
  • 16. The game is played by wandering around and finding your way home in a unique, dynamically generated environment each time.
  • 17. Stats like Vigor and Courage impact the player's speed and viewport size as they decrease.
  • 18. Higher resolution images could be achieved by adding an AI upscaler to the pipeline, but this would increase processing time.
  • 19. More creativity can be added to scene generation, such as user-selected themes or local weather conditions for immersive experiences.
  • 20. This process can be applied to other projects, taking advantage of AI-generated content.

Source: AI Engineer via YouTube

❓ What do you think? What are the most creative ways to leverage generative AI in game development, and how might these innovations transform the player experience? Feel free to share your thoughts in the comments!