📫 Subscribe | 🗣️ Hire us to speak | 🤝 Sponsor this newsletter | 🎓 Events

Welcome to the Friday edition of our newsletter. We spend Fridays going deeper into tools and trends related to generative AI (and Tuesdays sharing news updates). This week, Professor Snider marvels at ChatGPT’s new image model.

For the last five months, my default image generator has been Google Gemini's Nano Banana (we wrote about it here). Before that, it was Ideogram and Midjourney. Before that, the original ChatGPT image tool. It's been a constant rotation – whichever one was best at that particular time got the work.

It's time for another rotation.

OpenAI launched ChatGPT Images 2.0 on April 21, and after a week of testing it, I'm back to making my images in ChatGPT. It topped the LM Arena image generation leaderboard by 242 points – the largest lead anyone has ever recorded on that benchmark. The major change here is that this model thinks before it draws. It can not only make impressive photos, but also flawless infographics that will fool people into thinking they weren’t made with AI.

Here are three things to know about it – and a couple concerns at the end.

Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.

Try Wispr Flow free

Most image models treat your prompt like a bunch of keywords. They scan the words, pattern-match to training data, and crank out something close. ChatGPT Images 2.0 does something different. Before it generates anything, it plans the composition, considers spatial relationships, thinks about where text should go, and even searches the web if it needs current information.

You can watch this happen in the ChatGPT interface – the reasoning chain shows up before the image does. It's the same shift text models went through about a year ago when "thinking" modes started winning at complex tasks. Image generation just caught up.

If you've used AI image tools, you know the pain. You ask for a poster with the words "Spring Sale 2026" and you get "Spirng Sael 2026" or some made-up font that looks vaguely like letters. Even models that have figured that out get all the small text in the background incorrect.

ChatGPT Images 2.0 reportedly hits about 99% text accuracy – including in Chinese, Japanese, Korean, Hindi, and Bengali.

For anyone making social graphics, slide decks, or marketing assets, this is the change that matters most. You can finally generate a working ad creative without the need to fix the typography. And you can create up to eight images with consistent design features with just one prompt.

This was a pleasant surprise after months of seeing images with that same ChatGPT “look” to them. When you create something with Google's Nano Banana 2 or the previous ChatGPT model, you can usually tell. Same default fonts, same icon style, same slightly cartoonish vibe. ChatGPT Images 2.0 doesn't have those tells. The outputs look like a designer made them. Clean, restrained, not overdesigned.

Two important caveats.

1. The basic model is free for all ChatGPT users, but the "thinking" mode that does the reasoning, web search, and multiple-image generation is locked behind Plus, Pro, and Business subscriptions. So when you read about all the cool capabilities, know that the best stuff requires a paid plan.

2. It's slow. Each generation takes 40 seconds to over a minute, sometimes longer with thinking mode on. Google's Nano Banana 2 finishes in 20-25 seconds. For one-off marketing assets, the wait is fine. For creating a lot of images, it's painful. Plan accordingly.

ChatGPT Images 2.0 takes us one step close to producing client-ready work without a designer touching it afterward. That's a meaningful line to cross. If you're currently relying on Canva templates for your design – it's worth a serious look this week.

I showed this to my visual communication students with one clear message - the tools are going to continue to change (this class used to use Photoshop and now uses Canva), but their fundamental understanding of design will still matter.

And we may have just found a new to use in class.