AI RankingAI Adoption Rankings
Third-partyHacker News ·2026-11-12 ·5 pts

Making a Game Boy Advance Game with GPT-6 Astra

Photo: Kataraloveswater / CC BY-SA 4.0 · Wikimedia Commons (cropped to 560x315)

It's my first blog post out there, so excuse my typos, I'm writing all of this by hand.

Do you remember that "Bowman" flash game? It's a simple archery game where the goal is just to hit your opponent. I played it a lot when I was a kid. Now that GPT-6 Astra is out, can it build a polished remake while following my guidance on art direction, game loop, and game feel.

The GBA is a handheld game console that was released in 2001. I remember how badly I wanted one when I was a kid. I played on a Game Boy Micro for quite some time before getting a GBA. Playing gen 3 Pokémon on the GBA is still one of my favorite childhood gaming memories.

So, can an LLM in 2026 make a game for the GBA ? Let's find out.

My initial prompt set the core game loop, the overall art direction, and some gameplay details.

It was playable, but it lacked of FX and juice I expect from a polished game.

I used my own AI pixel art generator to generate the assets. I just gave the API docs to OpenAI Codex, and it figured out everything: Generating the assets in a consistent style, editing them when needed, animations etc.

There is also an MCP that connects directly with Sprite Fusion for ChatGPT/Codex and Claude.

The LLM figured out amazing variations for the leaves. The player and enemy characters were cool too. The grass animation needed a few retries, though.

I once again tweaked my prompt to add more juice to the game: smooth camera movements, a blood effect when hitting the opponent, bow recoil, arrow trails, hit feedback, and aded the animated grass and leaves.

Click inside the game to use the keyboard.

Download the Bowch! ROM (.zip)

Honestly ? We live in crazy, crazy times. It is just mind-boggling that a machine was able to code the whole game in plain C, compile it for the Game Boy Advance without any errors, and use all the tools (assets API, emulator tests, etc.) on its own to follow my art direction instructions.

It's easy to get a sloppy prototype from an LLM, but also surprisingly easy to make it great. The only difference is: Do you know how it should look and feel ? If you have no idea, prompting the LLM to "make it pop" wont save you.

Steering the LLM takes a good intuition for how the game should feel. How might we build such intuition ? I think it's by exposing ourselves much more to art and creativity (not only game-related art), which strengthens our ability to recognize why we like something.

Read the original www.spritefusion.com ↗

Headline, image and article body come from Hacker News; all rights remain with it. Reproduced unmodified for indexing — please refer to the original.

“GPT 6 Astra” in our rankings: View the boards →
Our data and derived results are licensed CC BY 4.0. Copyright of indexed items belongs to their respective sources.