I could have made this easier on myself. A subscription to one of the big image generators, type a few words, and a glossy picture within seconds. No install, no terminal, no wrestling with memory.
Yet I do it differently. My images come from a model on my own laptop, in one fixed style, with the occasional evening of frustration thrown in. Part 1 explained the why and part 2 the how. This part is about the choice itself.
The choice for one style
The design of this site is called “Press” internally. It leans on screenprint, coarse halftones, a limited palette, the look of something printed rather than rendered. When I went looking for imagery to match, sharp photorealistic pictures were not it. They fight the design instead of carrying it.
So I picked one risograph style and held it across everything. That is a constraint, and that is exactly why it works. A fixed style block, only the scene changes, the same seed. The result is that the images belong together without me having to reinvent it each time. Consistency beats spectacle, especially on a site people come back to.
This touches something I have also written about in photography. A view of your own is not a trick you switch on. It is a choice you keep making, until it runs by itself.
What it costs
Being honest about the price is part of it. The generating itself is not the problem. That takes seconds once the model is warm. The time sits in the toolbox around it.
The first few times I ran into all sorts of things. An environment that was half installed. A component that worked one time and not the next, depending on where I stood in the folders. A model that went silent instead of giving an error. And one expensive lesson: hard-killing a stuck process while it was using the graphics card cost me a full restart of the machine. Ever since, I stop everything cleanly.
That is not an advertisement for local work, it is the truth about it. Whoever chooses this also chooses the occasional evening of figuring out why something will not start. For me it is worth it. For someone who just wants a quick picture, honestly, it is not.
What it gives back
Against that stands something I no longer want to miss.
I own it. No meter, no subscription creeping up in price, no service that can decide tomorrow that my style is no longer allowed. The model sits on my disk and keeps working, even when the internet is down.
I recognize it. Every image on this site comes from the same hand, with the same choices. That is hard to copy, precisely because it is not a button but a habit built up over time.
And I learn from it. By doing it myself I understand what an image model can and cannot do. I carry that insight with me when I talk to others about this technology. You only really understand a tool once you have been stuck with it.
The wider lesson
This triptych was about pictures, but the lesson underneath is bigger. It is the same one that ran through my pieces on digital sovereignty and on leaving WordPress. As AI does more for us, the question of who keeps the control grows more important, not less.
Making images locally is a small exercise in that. Not out of fear of the cloud, and not because local is always better. But because it feels right to keep a part of your tooling genuinely in your own hands, to understand how it works, and to decide for yourself how it looks. Convenience is nice. Ownership is better.
This is part 3 of a triptych on making images locally with AI. See also part 1 and part 2.