The image generator turns your words into a picture. It runs the same trick as everything else in this series, pointed at a different job, and it is the corner of AI carrying the most anger, for a reason worth understanding.
Second of the three quick looks. Same deal as the last one, close enough to see how it moves, then back out. This one is lighter, because you do not use it the way you use the chatbot, and the point is only to know what it is when you meet it.
Words in, a picture out
You type a description and it paints exactly that, out of nothing, in a few seconds. Underneath it is the pattern-learning from the first piece all over again, only this model learned from a mountain of images with their descriptions attached instead of a mountain of text. See enough pictures labeled sunset and it works out for itself what a sunset looks like, closely enough to build a fresh one on request. The tell that you are looking at an image generator and not a chatbot is the swap. Words go in, and something to look at comes out, not something to read. That same idea has spread past still pictures into video and music, so the clip and the track are the same move wearing new clothes.
The first time I used one it did not feel like software. I typed a sentence, the kind you would use to describe a scene to a friend, and a few seconds later the scene existed, lit and framed like someone had made it. Two feelings arrived at once, the wonder at what I was looking at and the knowledge of where all that skill was learned. Keep both. They are both true, and this whole tool lives in the space between them.
The fight underneath it
These are the models carrying the anger from the piece on where AI came from, and it helps to know why. These models learned by studying enormous piles of art and photographs pulled off the internet, most of it made by working artists who were never asked and never paid. To a lot of those artists it feels like the tool competing for their job was built out of their own work, handed over without consent. That fight is still running, and nobody can tell you yet how it ends.
I am naming it, not weighing it. Whether that is theft or fair game, and what any of it means for artists and the rest of us, is a real question and it gets its own place later in a series built for it. Here I only want you to see the fault line, because it explains the heat around image generators better than anything technical does. Worth knowing too, one tool answers the fight head on. Adobe Firefly trained only on work it had the rights to, which is its whole pitch, the image generator you can use without standing on that fault line.
The core of it
Forget the product names, same as last time. Midjourney leads on looks today, something else will tomorrow, and none of the logos are worth memorizing. How exactly these models build up an image I am leaving alone here on purpose. Remember just this. It learned from other people’s pictures to make new ones from your words, and that is exactly why it is this useful and this contested at the same time. More than that you do not need here.
Next, the last of the three, and it is different from the other two. Everything so far only talks. This one acts.
Next up: A Closer Look at Agents →