Everything You Do Wrong
On craft, imperfection, and why the machine can copy the mark but not the reason for making it.
Frazer Irving will tell you, with the quiet amusement of someone who has long since stopped needing to impress anyone, that your style is everything you do wrong. He attributes it to Jerry Garcia, loosely, the way you’d cite something that turned out to be true before you’d even properly verified it. He says it almost as an aside, leaning into a thought rather than landing a speech, which is exactly how Frazer tends to make his most useful points, and it unlocks something that the entire current argument about generative Ai and creative work seems determined to avoid.
He is talking about deadline pressure, and about the last hours on a page when perfection is off the table and all you can do is make decisions. What you choose to simplify. What you exaggerate. What you decide not to draw at all. Those choices, he says, are not flaws to be corrected, they are signatures forming in real time. “The pages people respond to most strongly are often the ones made under exactly those conditions, not the polished ones, but the ones where I stopped performing and just got the job done.” The contrariness in that observation is very Frazer. He has spent a career doing things the industry did not expect, from 2000AD to Batman and Robin to collaborations with Grant Morrison that pushed atmosphere and psychological unease into the mainstream without asking permission, and he has never described any of it as the result of getting things right.
What he is describing (though he would not put it this way) is authorship. The decisions you make when there is no time to second-guess them are the truest account of how you see. That is where the self leaks through, and it is also, not coincidentally, where the machines fall down.
One of Frazer’s sharpest observations about current Ai output is one of the simplest: “It is too good. Too consistent. Too evenly distributed in its attention across a page. It does not know where to be lazy.” Humans, by contrast, are constantly making value judgements, pouring effort into faces and hands and letting backgrounds dissolve, implying rather than stating, trusting the reader to meet them halfway. That trust is the foundation of comics as a medium, and it lives in the decision about what to leave out rather than what to put in. He comes back to the gutter, the space between panels where the reader’s brain does the real work, time, motion, emotion, all of it happening in the gap. Films show you everything. Comics make you imagine it. Remove that absence and you do not improve the medium. You destroy its participation.
He jokes that if someone truly wanted machine-made art to feel human, they would have to work incredibly hard to fake mistakes, to introduce awkwardness and inconsistency and hesitation. And once you are doing that, you have already conceded the argument. You have admitted that imperfection is the point.
When I speak to Ram V he arrives at the same place from a different direction. He came to comics as a chemical engineer who found his way to San Diego’s green rooms by making work people wanted, not by clocking decades in a queue. He thinks about the industry architecturally, about structures and succession and what happens when you remove a load-bearing wall. He is five out of ten on the fear scale, he will tell you, and he means it, the art form will outlast the software, something feral always survives in the corner refusing to be replaced. But the middle worries him considerably more than the peaks.
“People talk about Ai like it democratises the workforce,” he said, and then he took the word apart. Democratisation only means something if the person making the thing is the one who benefits. If the machines strip value from the middle, the place where most of us work and most of the culture actually gets made, then what we are calling democratisation is erasure. The Otomos and the Moores will command attention regardless. The people on the rungs below them will find the ladder pulled away.
The on-ramp is what concerns him most, because most of us begin by copying someone we love, spending years sounding like someone else while our own voice surfaces underneath the imitation. A thousand hours used to be the price of entry, but if a prompt hands you a pass, you never find out whether you wanted it enough to keep going when no one was watching. He has seen the party trick where someone feeds a model asking for a script in the style of Ram V. It formats like a professional, packages itself with confidence, tells you it has humanist concerns and a shadow leaning toward horror. It understands the brochure of a story, not the story. He is not naive about how good the fine-tuning will get. The danger, he said, is “Not that it cannot copy him. It is that copying him freezes him.” A model learns a snapshot of a voice. The answer is evolution. Change outfits. Move the key while they are still clapping for the last song.
Then he told me the story about Otomo.
One of Katsuhiro Otomo’s assistants watched him ink the black dome that swallows Neo-Tokyo in Akira. It is, at its most reduced, just a shape. A mass of dark. You could achieve something functionally identical with a bucket fill and be done in seconds. Otomo crosshatched every square millimetre by hand until there was no white left. When his assistant asked why, he said: “There are a billion people dying under that dome. I have to count.”
That is the thing the current conversation about generative Ai almost completely fails to account for. Not the output. The reason. The moral weight carried inside a gesture that, to any outside observer, looks like a grown man filling in a circle with a pen. Software will never count the dead inside a gesture. It can fake the pattern. It cannot fake the reason.
Frazer would recognise that instinct immediately as he talks with me about legacy, about what happens when great artists are gone and there will never be another piece of work from their hand. He imagines a future where technology might preserve not just an artist’s output but their way of thinking, not through indiscriminate scraping, but through something focused, intentional, and consented to. He references a science fiction story about a headset that lets someone paint like a legendary artist by downloading their mind, with the catch that the user starts to inherit the artist’s madness alongside the technique. He laughs at that, but the point lands. Art is not separable from neurosis, instinct, obsession, and failure. Any system that pretends otherwise is lying to itself.
What he does see as genuinely useful is something narrower and more honest than the current pitch. Not a general model trained on everyone’s work without asking, but a personal tool trained on you. Your taste, your decisions, your mistakes. Something that reduces the grind while leaving judgement, meaning, and authorship intact. An assistant, not an author. The history of art is full of workshop models, Renaissance studios, Japanese manga production houses, teams built around a central vision. The ethical version of this future looks like that. The machine serves the artist. Not the other way around.
Ram would add that the punk side of comics has never needed anyone’s permission to survive. Pen, paper, stapler. A table at a con. A room in Yorkshire. He thinks a second indie boom is possible, driven by the artefacts of real work, the messy notebooks, the scratched margins, the flaws that carry fingerprints. Scarcity will do what ethics has failed to do. The work that refuses the machine will become more precious for its refusal. It is not the victory either of us wanted. But it is, as he put it, a map on how we might walk toward a logical conclusion.
What these two conversations share is not a position on Ai so much as a shared understanding of what craft actually is. Frazer approaches it through process, through the mark and the decision and the error that becomes a signature. Ram approaches it through structure, through the ladder of succession and what the industry loses when the middle collapses. But they converge on the same point. That what generative Ai threatens is not output. The machines can produce output indefinitely. What they threaten is the long, costly, necessary process through which a person becomes themselves, the accumulated wrong turns that eventually add up to a voice no one else has.
Otomo counting the dead inside a dome of ink. Frazer making a deadline decision that leaves a background unfinished and trusts the reader to complete it. Ram writing the draft that still sounds like someone else, because the next one will finally sound like him.
None of that is efficient. None of it can be prompted. All of it is what the work actually is.
The machine will learn to fill the black fast. The rest of us will have to count the bodies under the dome.
Drawn to Extinction is out now. If any of this landed, the book goes much deeper, into the voices, the history, the law, and the people holding the line. You can find it at your local independent bookshop, on Amazon, or at comicvault.cc.