X (formerly Twitter) ·
Presenting - chatGP-T1000 (our submission to
@AssemblyAIhack-a-thon) An AI shape-shifer bot, that assumes identities understands your natural language, and replies with chatGPT responses, in character with deepfaked audio and lipSync. Big thread with tons of examples 🧵 LFG
First - acknowledgements. We've rebuilt the code from scratch for this hack-a-thon, using the awesome chatgptpy
@RawandShaswar👏 Voices are generated by
@__UBERDUCK__and Whisper understands you. We also found wav2lip and are generating "speaking videos" with it. Now demo:
You can join this TG group and see all the progress we made (and maybe if you're lucky, interact with the bot!)
My kids LOVE lion king, and were REALLY excited to help me test this new thing. Here's my daughter talking to her favorite characters from the movie.
How about a personalized rap from Vanilla Ice about the current state of the economy? 🤣
Since this was a hack-a-thon, and openAi doesn't have an official chatGPT API, this bot will not work publically, and will likely break until an official API launches. But it was SO fun to work on! Check out our submission on devpost and give us likes!
Here's a UML diagram of how this works: Again, all the code was written from scratch, during this weekend for
@AssemblyAIhackathon. And it seems that now exactly 1 hour after submission the bot is no longer live due to chatGPT catching up. 😅 But we made it to submission!
A very interesting challenge was to deepfake the bot resposes, I've tried asking twitter for this, and came up short, and we used wav2lip eventually, which only animates the lips
I really want to do another shoutout to
@__UBERDUCK__Their service is amazing, super super quick, and voices sound very decent! Thanks UberDuck! This wasn't possible without yall
