End up building a decent ADV game engine with Antigravity and Chat GPT

Source: Dev.to

Introduction

I didn’t expect this at all – at first I thought I could use Antigravity just to set up a Phaser 3 project and then code everything manually.
It turns out I haven’t written any code yet; the only thing I do is provide scenario/image data, and I’m continually amazed by how much the AI understands and how capable it is. I have very specific needs for my game design (Phaser 3 already has many capabilities), and based on my past experience I had a general idea of how to structure my scenario files. The AI extended that idea far beyond what I imagined.

Original Scenario Format

My initial format was a CSV file with headers like:

scene,idx,op,speaker,message,text_id,bg,char1,char2,char3,icon,ui,tags,goto

If you’ve ever made an ADV game you can guess what these columns represent. I was happy to externalize scenario data from the main game codebase, allowing quick iteration, fixes, and testing. However, CSV isn’t the easiest format for humans to edit, so I asked Chat GPT for advice.

AI‑Suggested Workflow

Chat GPT suggested using Antigravity to write a Node utility that reads a much simpler “raw scenario” format and converts it to the CSV structure. After a brief explanation, Antigravity produced a one‑shot implementation (with only minor tweaks needed).

Raw Scenario Example (opening_jp.txt)

@bg destruction_of_world
@ui DIALOG
@chars alexis@C:neutral
@face alexis:neutral
アクシス：な、何が起きてるんだ？

@bg black
@ui NODIALOG
@chars clear
@face clear
[NARR] それは、もう終わったはずの物語だった

Converted CSV (scene_script.csv)

opening,10,SAY,アクシス,な、何が起きてるんだ？,,destruction_of_world,alexis@C:neutral,,,alexis:neutral,DIALOG,,
opening,20,NARR,,それは、もう終わったはずの物語だった,,black,,,,,NODIALOG,,

(I’m building this game in Japanese, but the structure should be clear.)

In‑Game Result

The converted data is displayed in the game as shown below:

Note: This is still an early prototype and will evolve. The goal right now is to display characters, show face icons, and present dialogue.

Conclusion

Combined with a preliminary yet powerful event system via events.json, this workflow lets me focus on writing the story rather than plumbing. I’ve spent roughly one day on the setup, which feels like a huge productivity boost compared to my previous Android‑game development experience—all thanks to AI!