It will come as no surprise that AI image generation has come a long way in the last three years. I remember how excited I was when I first began playing with MidJourney not long after it had launched as a Discord bot in February 2022. I’m a bit of a digital imaging nerd, so I was giddy.
The earliest sample I can find is from July 15, 2022. It is shown above at the only size possible at the time. I gave Midjourney a straightforward prompt, “a plant with human heads as leaves.” The resulting image was 256 x 256 pixels, and the only option was a 1:1 aspect ratio. At the time, this felt like some form of wizardry.
Below is what Midjourney pumps out with that same simple prompt today. The base image was 1048 x 1048 pixels, and I told it to build it at 1:1 for comparison, but I could have set any aspect ratio. And I used the “creative upscale” feature to take it to 2048 x 2048 pixels, then resized it in Adobe Lightroom for display here. This is a good illustration of just how far Midjourney has come in just over three years. It still feels like some form of wizardry.
AI Image Generation Has Many Players Now
For what felt like a long time, Midjourney was my “go-to” AI image generation tool. There were an are numerous others, but it was my favorite until just recently. Most of what I was using it for was slide backgrounds, and it was and still is excellent for illustrating abstract concepts. I used it for the thumbnail image for this post, for example.
What it has never been great at, however, is text. It’s still not good.
In the last several weeks, the leaps made by ChatGPT and Google are massive. It’s mind-blowing how well these tools create images. This includes accurately rendering text in just about any style you can imagine. And I’m going to use a logo design to illustrate just how far this has come.
My wife goes to an excellent barber to get her hair cut. His fades are tight. In December of 2022, he was considering leaving the shop where he worked and joining a few of his friends to start a new venture they would call “Billionaire Barber Club.” When she told me about it, I decided to see what Midjourney would do with a prompt like this: “a logo for a barber shop called “Billionaire Barber Club,” barber scissors, flat art, vector art, solid colors, no shading.” Here is one of the results.
While it could never pass for a “finished” logo, it was suitable for ideation. I played with several different prompts, and the results were interesting, but nothing was really usable without a great deal of work in Adobe Illustrator. In the end, he didn’t follow his friends, and they went a traditional route for the logo design. But when I fed the same prompt to Midjourney today, the designs were wildly different, but it still struggled mightily with text. Again, great for quickly generating lots of visual ideas.
AI Image Generation in ChatGPT & Gemini
Today, I fed that same prompt to ChatGPT Images and Nano Banana Pro in Google Gemini. I didn’t alter the prompt, though I could have iterated to even better results, I’m certain. As you can see from the samples below, we’re not even living in the same world we were living in just three years ago. I’m going to let the images speak for themselves for a moment.
“I Am Oz, The Great And Powerful”
“Any sufficiently advanced technology is indistinguishable from magic” is the Third Law of Arthur C. Clarke. That’s the way this all feels. As if we’ve entered the Emerald City, and all we need to do is ask, and the great wizard of AI image generation delivers.
I could give you hundreds of examples, from hyper-realistic renderings of West Virginia creeks to nearly instant image alterations that would have taken many hours in Adobe Photoshop just three years ago. Thankfully, these platforms have allowed me to archive early attempts across a wide range of styles and formats. It certainly makes it easy to demonstrate how far we’ve come.
And we’re certainly not in Kansas anymore.
Add your voice...