Short answer: Sam Nadler says reports point to much better text fidelity in ChatGPT Images 2.0 and he finds it pretty good. Jordan Metzner was underimpressed, saying the launch examples beat his results versus Nano Banana.
From Episode 40: Claude Design Changes Everything (Figma in Trouble?), starting at 14:10.
Sam Nadler [14:10]: I'm excited to see what will happen. Obviously, you can't count Elon out, but it's a bold move. That's for sure. The other article we're going to cover today is yesterday, ChatGPT launched Images two point zero. I know we've both had very limited time to play with Images two point zero. I think your reaction has been underwhelming, while my reaction has been, it's actually pretty good. I think what I've heard, mainly on social media, but according to this article as well, is that the text fidelity is really good. In previous models, you would write something like, Jordan.
Jordan Metzner [14:46]: You'd get, like, gibberish. You'd get gibberish. Yeah.
Sam Nadler [14:48]: Yeah. You'd j r d n or something like that. Mhmm. Mhmm. But apparently, for this, it's, like, really high text fidelity, which is, exciting.
Jordan Metzner [14:57]: Yeah. I've been playing with it a little bit. I still I thought I was a little underimpressed, to be honest. I think some of the examples they showed were more impressive than the results I was getting. This background that I generated today was generated using Image two. And, on almost every episode, Sam, I generate a artificial background, a new one. I don't know if there's a significant difference between, like, my previous backgrounds that were being generated by Nano Banana or not. But, yeah, overall, it's interesting to see ChatGPT into the space and get deeper and compete against Nano Banana, where it seemed like Google had almost, like, run away with it a little bit for a while. But, yeah, it's interesting, especially because they shut down kind of the Sora business of, like, generating content. So it is a fun, cool space.