Launchcaster!

Excited to share an early VR beta of http://Meshi.fun on Sidequest!

Try it here: https://sidequestvr.com/app/17400/meshifun-ai-characters-in-vr

Meshi lets you talk to 3D AI characters. Characters learn as you talk, and anyone can create a new character!

Let me and @kerls know what you think :)
🧵⏬

Currently between things, prev @ plaid, south park commons, caltech and did other things - omarmezenner.com

We're both really excited for the prospect of AI games, but as a small & mighty team of 2 we had to stay focused on our core game loop. We weren't as excited about "just using AI to speed up game dev". 

We wanted a way to empower players to create with AI, and landed on creating & talking to characters

2) Cost kills you every step of the way. For a fully engaged user, we pay $10-15/hr, split about 1/3 to openAI and 2/3 to eleven labs
.

We experimented with a Twitch stream format with a multi-user queue on our backend so people "share" tenancy. But most ppl wanted 1:1 convos

3) A good conversation takes 2! You really do need to "juice up" character personalities to drive the conversation. Our initial character was the lovable Inuinu, who we made extra tsundere. Some people were turned off by its rudeness, but most loved it!

If you're down to clown with us, join our Discord! We're always looking for testers for new features: https://discord.gg/RQ2jZ2tU

Challenges!

1) Latency kills you every step of the way. 3-5 secs for LLM responses (openAI), 2-4 secs for TTS (eleven labs) , 500-1000ms for STT (azure) . It all adds up, and conversations can feel sluggish. We batch incoming messages, but initial latency still suffers.