I use AI for reference, not drawing

Let me be specific about what I actually do, because the word "AI" gets used to describe everything from "I used Photoshop's generative fill once" to "the AI makes the image and I write a prompt." My usage is neither.

What I use it for

Lighting reference. Texture generation. Color palette exploration.

That's mostly it. When I'm working on a complex scene — say, a figure in a specific lighting environment I haven't drawn before — I'll run a few quick image generations to get a sense of how the light would actually fall. I take a screenshot. I use it as a reference while I ink, the same way I'd use a photo.

For textures, I generate paper, fabric, and surface textures that I composite into finished pieces. Not as a replacement for hand-drawn texture — I still do linework by hand — but as a layer underneath that adds tactility without the time cost of rendering every surface manually.

For color, I'll sometimes generate a version of a finished piece in the color palette I'm considering, purely as a preview. It takes three minutes and tells me whether a decision I'd otherwise spend an hour on is going in the right direction.

What I don't use it for

Drawing. Composition. Linework. Character design. Expression. The thing that makes a piece a piece rather than a render.

I made a deliberate decision early on to keep the generative tools out of the parts of the process where the human judgment is what creates value. If AI could do what I do with a brush pen on Bristol, the work wouldn't be worth collecting. Since it can't — not in the ways that matter — I use it for the things it genuinely does faster and better: variation, reference generation, texture.

Training a LoRA on my own work

I've done this. The intention was to speed up the reference stage further — instead of a generic model, a model that knows my style and can generate lighting references that already roughly match my aesthetic. It works, with caveats. The model knows my line weight and my palette. It does not know why I make compositional decisions. The outputs are useful the same way a photographic reference is useful: as a starting point for a human judgment call, not as a finished answer.

The honest version of "AI as a tool"

Everyone says it's just a tool. Most people saying that are either defending heavy AI usage or apologizing for any usage at all. The reality is more boring: I have a workflow that produces work I'm proud of, AI sits in specific parts of it where it saves time without compromising the result, and I've drawn a line at the parts where compromising the result would be the cost of using it.

That line might move as the tools develop. I'm watching closely.