A 1st Person Point&Click Adventure, created with Godot and AI

In a nutshell, this is about "Godot + AI + Me creating a 1st Person Point&Click Game Prototype".

Inspired by Tim Rachor's great Zarathustra game, I created a 1st Person Point&Click game as vertical slice. The main goal was to use as much AI assistance as needed (not: as much as possible). Another goal was to see how much effort this type of game would require from me.

So, here we now have "The great Toilet Escape" game (working title: "Escape from Pipi-Kaka-Land", but I was told that sounds a bit too weird...)

Gameplay

The gameplay is intentionally simple: The player is trapped in a toilet room and has to escape (i.e., open the locked door). To do so, he/she must take the gloves from the cabinet, use them to retrieve a key from the toilet bowl, and then unlock the room's door. There is only one game scene. Voila. Fun, thrills and excitement. Not 😁

AI Support

As AI should help me with this project in various manners, I got

the lowest paid ChatGPT Plan (to avoid long waiting times)
LM Studio, locally installed (for LLM usage)
Fooocus, locally installed (for image generation)

Pretty soon, it turned out that local AI is easy to install and is actually working - many results are questionable, though. At least on my rather old PC. So, ChatGPT (Web) was the main way to go for me here. Also, it could be used for both text interaction and image generation (GPT-5.5, June 2026).

AI Tasks

I used AI support for

gameplay ideas and design
creating graphics (a lot, uh)
UI design
software architecture improvements
code smell analysis
dialogue text generation
general topics like using Godot, programming gdscript, etc

The main architecture and programming of the game was done by me - simply because I wanted to. I actually learned a lot of new things through proposals given by the AI - but did not implement changes I didn't understand. This is a key thing when working with AI support, imho.

Also note that the goal of this project was not to try to create a complete game fully via prompting. I could not learn anything from that, I think. So, why do it then in the first place...

UI Interface

The game does not feature a walking avatar (as being a "1st Person Point&Click"). The inventory system is very basic, too (no "combine things" screen). After all, there are only 3 inventory slots used, anyway.

Game States

Well, not really exciting. This image is also AI generated. I should have done it myself, though...

Reaction Definitions

Each interactable item in the scene must have a "view" interaction (eye sign). Some only have this (e.g. a wet towel on the wall), some do also have a "grab" interaction (hand sign). Depending on the current game state and inventory, these interactions have different reactions and text lines assigned.

To manage different reactions based on input combinations, I used an Excel file, filled with data and exported to *.csv. This csv is shipped with the game and loaded and interpreted during runtime. Based on this "decision tree", outputs and reactions are managed ingame.
This solution only has limited scalability, though (yeah, I hear you, AI consultant, I hear you).

Tools Used

Godot 4.6.3, Paint Shop Pro5 (antiques ftw), ChatGPT GPT-5.5 (Web), freesound.org, Excel

Final Words

Using AI enabled me to realize a project I otherwise would not have been able to create (mainly, because of the graphics). As long as one checks AI recommendations and understands them(!) before implementing, it is a great learning experience, too.

Downloads/Links

https://www.mikoweb.eu/godot/efpkl/
Link to the (debug) Godot Web Export of the game (fixed 1600x900px resolution, no adaption to browser window size)

efpkl.zip, 8 MB, 12.6.2026
The source code of the project (Godot 4.6.3)