GPTAvatar: I made a creepy AI virtual avatar version of myself! (Unity, ChatGTP, elevenlabs)
So what happens if you take all the hottest AI APIs out there and combine them into a single Unity app? You get to bully a virtual version of yourself!
This is kind of weird, but it really shows how good AI stuff is these days, especially voice synthesis. With some of its dialog, I don't think you can tell it isn't really me speaking.
The first part is me showing Akiko the app on an iPad, then the next part is me just playing around talking to myself and trying to piss me off, basically just showing the strengths and weaknesses of how it works.
To be clear, the AI Seth is creating and speaking on the fly, nothing is pre-recorded or scripted.
Part of his knowledge/behavior is from my prompts (he's told to pretend to be me), but most of it is coming from ChatGTP. That's how how he knows everything about every game in the world (including my stuff on wikipedia at least) - but he also tends to lie a lot and misrepresent my real views.
Has technology gone too far? Why would anyone want this?!
Lots of ways to improve this, one would be the character should be making eye contact with the camera, and better head/jaw rig, the way the mouth moves is pretty bad...
You can download this app or its source code here: https://github.com/SethRobinson/GPTAvatar
Keep in mind to use it you need API keys for a bunch of stuff, and using those APIs can really cost a lot so beware.
What I used to make this:
Engine: Unity
Listening: Whisper (via OpenAI's API)
Thinking: ChatGTP (via OpenAI's API)
Talking: ElevenLabs' Voicelab (trained on my voice)
3D Model: Reallusion's Headshot (using my picture)
Lipsync: SALSA Lipsync Suite