RealTalk Returns: We Just Created The Deepest Deepfake Yet. Here’s Why.


Ladies & Gentlemen, welcome to another episode of ‘I’m not Joe Rogan’. No, seriously. I am not. This is Joe Fauxgan speaking an artificial intelligence created by Dessa. You guys know David Barstow? He’s a New York Times reporter who won the most Pulitzer Prizes for journalism in history. He recently followed a crazy deep learning company called Dessa, as they created the world’s first combination of realistic AI voice and video. Here’s what they created… I’ve been thinking, and I realized, that it’s been almost ten years since the first episode of the podcast, that’s f***ing crazy. This show has become my life. That’s why I decided to go out with a bang. So, on December 24, we’ll be doing our last episode, ever. That’s exactly ten years since the first episode came out. It’s been a good run and we’re all going on on top.

45 comments

To improve the facial result at 0:30, don't use the seamless mode. Instead, use the overlay mode with color transfer set to LCT or RCT.

Voice is great, face still a little stiff. Just watched the tv special. I’m curious how much further this could be pushed with hand done editing in post. I wanna see it done completely, so that everything is forced to be questioned

This will collapsed democracy. Democracy is built on public trust and accountability. With this technology there will be no accountability and thus no democracy can survive.

2020 election's gonna be wild, who's ready for some last minute election night deepfakes?! WHOO!!!

Make a Tai Lopez one, so we can make Joe Rogan and him talk about Lamborghinis versus Tesla. Stoned apes would have dropped out of college or reading a book a day on DMT.

Crazy how as humans we can till that's it's still not quite the real thing. Choppy speech skin tone not correct, no natural fuck ups in then
speech

My gosh man. This is insane. There are ways you can tell it is somewhat a super imposed face, but the voice is uncannily real. I wonder what Joe Rogan will have to say about this? I wonder how long it took to produce this?

The visual is honestly run-of-the-mill by today's standards, but the audio….it sounds like he's reading a script, but that's just it. It sounds like he is reading the script. When the software gets better at emotional intonations…

Would that be feasible to redub a bit in a different language? I am aware that flow of emotion and information is different for each language, but would your method work for cloning voice to a different language speaker, as to be used with translation or is it limited to one language, or perhaps just English? Would be really interested to see this scene done 🙂
www.reddit.com/r/Norway/comments/cv16c6/vinland_saga_scene_dubbed_in_old_norse_episode_7/

Good effort although the lighting looks a bit baked and the spec doesn't look right like I'm looking at photos projected on a 3d face rather than lit and rendered correctly. Which I suppose is what deepfake is.

Leave a Reply