I’m a big fan of AI. It’s changing our world at an unprecedented pace, and I’m pretty sure it has the chance to propel humanity into a bright future. But that will be the topic of another blog post, in which I explain why I’m an AI optimist.
Last week, Midjourney released version 5 of their image generation algorithm. One of the very first things I try with every new image generation is to ask it to generate a band photo. I have some background in music photography, and it also seems like a good way to test some things whether it’s good at composition, lightning, humans and famous people.
Until now, the results were terrible. Take for example this prompt:
“an award-winning photo portrait of Amy Winehouse taken in a dark & smokey bar, great lightning, realistic skin”.
You couldn’t let anyone believe this was really Amy Winehouse. Some results look a bit like her, but the uncanny valley is pretty deep. And none of these pictures would be award-winning.
Now, if you ask Midjourney 5 the exact same prompt:
Now, this is a good portrait. I can feel it. The hand is still a bit weird, but if I saw this picture in a magazine, I wouldn’t probably notice.
Playing around with Midjourney 5 gave me the idea to collect these portraits on a site. In part because it’s a fun little project, but also because these algorithms are getting better and better, and I want to track the progress.