Image Generation Edition
LAST UPDATED: April 26, 2026 at 11:39 AM
by Braden Kelley
Watching the evolution of AI over the past eighty years (83 actually) has been fascinating to watch (admittedly, I haven’t been alive long enough to watch all of it), but the evolution over the past 3 1/2 years following an extended AI winter has been nothing short of amazing. To anchor us and set context for what’s next, here is ChatGPT’s evolution over the current AI spring:
The Evolution of GPT Models
A quick reference for the major milestones in generative AI development:
| Version | Release Date | Key Achievement |
|---|---|---|
| GPT-3 | June 2020 | The first massive 175-billion parameter model. |
| ChatGPT | Nov 2022 | Brought generative AI to the general public via a chat interface. |
| GPT-4 | March 2023 | Introduced advanced reasoning and multimodal (image) support. |
| GPT-5 | August 2025 | A “network of models” approach for complex problem-solving. |
| GPT-5.5 | April 2026 | Current state-of-the-art model for nuanced reasoning. |
Earlier this week OpenAI released a new image model and people were wondering why, after killing of their video model Sora to focus their limited resources, would they introduce a new, potentially resource hungry image model that will burn more of their compute?
My uninformed user perspective is that perhaps OpenAI’s leaders saw what it could do and they just couldn’t justify depriving the public of it given their stated mission to “ensure artificial general intelligence (AGI) benefits all of humanity.”
I’ve created more than 1,200 quote posters over the past few years for people to use in their meetings, presentations, keynotes and workshops (download them for FREE at http://misterinnovation.com) using freely available images initially from sites like Pixabay, Unsplash, Pexels and Wikimedia Commons like the one above because the image generation capabilities of the AI models were so bad.
Then about eight months ago when Google launched Nano Banana the AI image generation started to be good enough at capturing the essence of a quote to use an AI generated image instead of a photo (see the example above), before layering the quote in a translucent layer on top of it.
But then in March 2026 I started using Gemini’s Nano Banana 2 to start creating hand drawn style images for the quote posters (like the one above) because of it’s ability to MUCH BETTER handle the inclusion of text into an image. You can see in this image, not only was it able to include the quote in the image, but it was able to add some other supplementary text (on its own) into the image AND an image of me, without me asking it to!
I started using this hand drawn style for many of the quote posters I’ve created over the past couple of months, doing a daily bake-off between Gemini, ChatGPT and Grok (which loses 99% of the time) and in March 2026 Gemini was winning most of the bake-offs until maybe April when it started to be about 50-50 between Gemini and ChatGPT.
BUT, with the release of OpenAI’s new image model earlier this week, ChatGPT has been winning every day and it is because it has been creating images like this one off a single, simple text prompt with the quote, author and requested style provided:
Now remember, all I gave ChatGPT was the quote and the author and asked it to capture the essence of the quote in a hand-drawn style. IT decided to add all of these other informational, education, inspirational elements and my jaw literally dropped.
If I was an OpenAI executive and saw this result to my prompt, I too would have argued for the release of this image model given OpenAI’s mission. This ability is superhuman. I as a human would have stopped at finding an image that reinforces or enhances the meaning of the quote.
This image model turned the quote into a multi-dimensional learning tool that transmits far more insight and information in a single document than the already powerful single sentence did.
The quote is still an important distillation that is far easier to remember and thus to drive behavior change from, but the rest of the content that the OpenAI image model created of its own volition adds value for those who want to quickly double-click on the essence and learn more.
So, this is where we are with AI image generation now, this is the kind of power these tools now have. The only question is:
What are you going to do with them next?
Image credits: Google Gemini and http://misterinnovation.com (download all 1,200+ FREE)
Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.