Junk Wars: AI-Powered Game Development in 10 Days
Key Concepts:
- AI-assisted game development
- Vibe coding (using AI for 80%+ of code)
- 3JS (JavaScript 3D library)
- Cursor (AI-powered code editor)
- Midjourney (text-to-image AI)
- Cloudflare Durable Objects (serverless platform)
- Trellis (2D-to-3D model AI)
- Game optimization
- Multiplayer game architecture
- Zuno (AI music generator)
- 11 Labs (AI sound effects)
Vibe Jam Challenge and Game Concept
The video details the creation of "Junk Wars," a browser-based multiplayer deathmatch game developed in 10 days for the Vibe Jam coding competition. The competition, announced by Peter Levels, required at least 80% of the code to be written by AI (vibe coding). The game needed to be instantly playable in a browser, have mobile support, and ideally include multiplayer. Judges included Andre Karpathy (coined "vibe coding") and Mr. Doob (creator of 3JS). The prize was $10,000. The game concept was inspired by prop hunt games, but adapted into a deathmatch format set in a junkyard, which was ideal for AI-generated assets.
Day 1: Initial Progress with AI Tools
The primary tool used was Cursor, an AI-powered code editor using models like Claude Sonnet and Google's Gemini. Cursor was used to generate 3JS code based on plain English requests. For example, the initial request was to "create a basic 3JS game that implements some of the core functionality here Please keep it really minimal and start off with the player being just the cube that can move around and can fire when I do a left click". Midjourney, a text-to-image AI, was used to create 2D junk item images, with Claude assisting in generating effective prompts. The images were then processed in Photoshop to remove backgrounds. Despite initial progress, the creator encountered challenges, including discarding 3,000 lines of AI-generated code and struggling with adding a walkable ground due to lacking the correct vocabulary for prompts. The term "heightmap" was the key to generating the map. By the end of day one, a basic game with a 3D map, visuals, and enemies was created.
Days 2-6: Core Features and Multiplayer Implementation
Day two focused on building core features like respawning, aiming, spawn protection, and transforming into junk. A single-player version was deployed on junkwars.net, and feedback was incorporated. The creator then tackled multiplayer, which required a complete re-architecture around a central server. The process took three days and involved debugging complex AI-generated code. At one point, five hours of work were lost due to a lack of version control. By day six, a local multiplayer version was functional and tested with the creator's children.
Day 7-8: Soundtrack and Deployment Challenges
A soundtrack was created using Zuno, an AI music generator. The creator spent a night editing, clipping, fading, and remastering the music. Day seven focused on deploying the game on the internet, which proved challenging. Cloudflare Durable Objects were explored, but the creator faced deployment issues and incurred $300 in server costs due to leaving servers running. By the next day, the game was live, and viewers joined to play.
Day 9-10: 3D Assets, Optimization, and Final Touches
The creator felt unfulfilled despite the game being playable. Feedback from friends led to the integration of 3D assets using Trellis, a 2D-to-3D model AI. However, the initial models were too large (e.g., 150,000 triangles for an AC unit), causing performance issues. The creator learned about optimizing 3D models. In the final stretch, 11 Labs was used to generate sound effects. The creator worked through the night, fixing bugs and adding features. The game was submitted hours before the competition closed after 36 hours of no sleep.
Judging and Final Result
After submission, the creator continued to troubleshoot issues. The game server crashed at one point. During the judging process, S13K, one of the judges, discovered a mobile compatibility issue on iPhone. The creator fixed the issue in 10 minutes, leading to a positive interaction with S13K, who played the game and provided valuable feedback. Junk Wars placed in the top five out of over a thousand games.
Conclusion
The creation of Junk Wars demonstrates the potential of AI in game development, particularly for rapid prototyping and asset generation. While AI tools like Cursor, Midjourney, Zuno, Trellis, and 11 Labs significantly accelerated the development process, the creator still needed strong coding fundamentals, problem-solving skills, and the ability to optimize and debug AI-generated content. The project also highlighted the challenges of multiplayer game development and the importance of version control and cloud service cost management.
AI summaries can miss context or contain errors. Check important details against the original video.