“Did the AI make this?” People ask it about my work like it's a gotcha, and I never quite know how to answer — because the question has a broken assumption baked in. Did the camera make Kubrick's films? Did the piano write the symphony? The tool is never where the art was hiding.
I use generative models all the time. I'm also completely uninterested in the version of this conversation that's a panic about them. The interesting question isn't whether the machine can generate an image. It obviously can. The interesting question is what happens in the thousand decisions around the generation — because that's where the work actually lives.
The tool was never where the art was
We've done this dance before, many times. Photography was going to kill painting. The synthesizer was going to kill “real” music. Digital brushes were going to kill illustration. Each time, the same arc: a new tool collapses the cost of producing a thing, everyone panics that craft is dead, and then — quietly — the tool gets absorbed, and the craft moves up a level to wherever the human judgement went.
Generative models are just the newest verse. They collapse the cost of the first draft to almost nothing. What they don't do — what no tool has ever done — is decide what's worth making, or why.
How I actually use it
In practice, a model is one instrument in a pipeline I'm directing, sitting right next to the 3D renderer and the colour grade. The loop looks like this: prompt, curate hard, composite, regrade, deliberately break something, and go around again. Generation is the start of the process, never the end of it. The model proposes; I dispose.
Sometimes I'm using it for an element I'll barely see in the final — a texture, a glow, a plate I'll paint over until it's unrecognisable. The output is raw material, not a finished good. Treating it as finished is exactly the mistake.
Technology is the brush. Innovation is the canvas. The hand is still yours.
Where the authorship lives
If the model does the generating, where's the authorship? In three places it can't reach.
Taste — knowing which of a hundred outputs is the one. Intent — knowing what the image is for, what it has to say, where it'll sit. Restraint — knowing what to throw away, and being willing to throw away almost all of it. None of those live in the prompt. They live in the years of looking that let you say, instantly, “not that one.”
The tyranny of infinite options
Here's the strange new problem. For most of history, making images was expensive, so scarcity did some of your editing for you — you only made what you could afford to make. Generation deletes that scarcity. Suddenly you can have ten thousand versions of anything by lunchtime.
And ten thousand options is not a gift. It's a fog. When everything is possible, the entire job becomes choosing, and choosing well is brutally hard. The craft didn't disappear; it migrated. It used to be in the making. Now a huge part of it is in the discarding.
Automation makes images. Authorship makes one image mean something.
Intent doesn't come in the box
The thing a model fundamentally lacks is a reason. It has no stake, no point of view, nothing it's trying to say. It will generate a masterpiece and a disaster with precisely equal enthusiasm and no idea which is which. Supplying the why — the intent that makes one of those images matter — is the human's entire job. It's the same principle I apply when I treat AI as one component in a system rather than the system itself, and the same post-production discipline I bring to directing a whole site.
I'm genuinely not worried about the brush getting smarter. A smarter brush is a better brush. What I care about — the only thing I've ever cared about — is what I can paint with it, and whether it's worth painting at all.
