Hi Guys,
In case you didn’t know about it, there is an interesting coding competition with some great prizes organized by Google. Create anything which uses Google’s Gemini AI API, in our case probably a JME based game. You have ~50 days to submit it.
Good luck!
4 Likes
So… more than a month passed and after a very intense effort, a first version of the AI based game creator is ready. It’s far from perfect but we have a complete story - the user can have a conversation with the Gemini AI agent and a game is created/modified/totally replaced in real-time until the user is satisfied. Finally per the user request, the AI will assemble a ready to deploy Android project with the source code and assets. There were a few challenges especially around the Android export (I had to downgrade to JME3.5.2 BTW) but now all is good and I have 3 more weeks to enhance the product and the presentation.
Some technical key points in this video:
- The conversation with the AI is automatically streamed to the server when the user completes its recent sentence this is to allow as much as possible natural conversation.
- The AI reacts to “non important” words like “thank you”
- The text-to-speech Lib is awful reminds me the speech quality of my Commodore64 back in the days
- The game changes itself in run-time. It doesn’t restart on every change but rather updates the code
- There is no 100% consistency. Sometimes the AI doesn’t react well to the user input. This is due to a relatively high temperature (creativity) set to the model.
Cheers!
11 Likes
Wow, this is awesome.
Good job with this.
1 Like
All projects were submitted up to Aug-12 and now the judges starting their rating. On Aug-26 it will open to public voting as well. I see that there is at least one quite similar project which is based on Unreal Engine. I think he did a good job and I’m expecting tough competition
4 Likes