If a client can train a LoRA on my back catalogue and generate "more in my style," what is an illustrator selling — images or authorship?

The shift

Producing an image in a specific, recognizable style — including a named illustrator's own style, cloned from a LoRA trained on their back catalogue — goes from scarce (that artist's hours) to abundant and near-free. The style, which used to be the illustrator's moat, is now the most copyable thing they have; what stays scarce is the living artist who authored it, can iterate on a brief, and can be held to the result.

The axioms

• An illustrator sells finished images, priced and turned around per deliverable. Rests on producing a polished image being scarce and slow, so the piece is the natural unit of sale.
• There is a commodity commission tier — spot art, icons, generic editorial, "something in this vibe" — that pays because even competent, unremarkable illustration needs a trained hand. Rests on "good enough" image-making being scarce.
• A recognizable personal style is a durable moat clients pay a premium to get. Rests on a style being expensive to imitate.
• The artist controls who gets to make work in their style. Rests on reproducing a style requiring the artist's own labour.
• A named illustrator is who a client trusts to interpret a brief, take direction, and stand behind the result. Rests on trust, judgment, and accountability being non-transferable and slow to earn.
• Iterating live on a brief — reading a vague ask, pushing back, revising toward what the client actually meant — is part of what's bought. Rests on responsive judgment under an underspecified brief being scarce.

Invalid axioms

1. An illustrator sells finished images, priced and turned around per deliverable. The image as a unit of scarce production is gone; output in a named style is generable in seconds at near-zero cost, and clients now arrive with their own AI mockups expecting matching price and turnaround. Habit-trap: pricing per piece, per asset, per revision-round still meters the thing that stopped being scarce, and it invites the client to treat the illustrator as a slower, dearer image faucet.
2. There is a commodity commission tier that pays because competent, unremarkable illustration needs a trained hand. This tier is the most exposed part of the market — spot art, icons, generic editorial, background filler are exactly what a model does well and cheaply, in any style asked for. Habit-trap: illustrators whose income sits here still treat volume commodity work as a reliable floor, when it's the first layer to be undercut to near-zero.
3. The artist controls who gets to make work in their style. Reproducing a style no longer requires the artist's labour — a LoRA on a public portfolio does it — so control-by-default-of-effort is gone. The belief was true only because imitation was expensive; the flip removes the friction, not the wish. (Whether the artist has legal standing over their style is a separate, unresolved question — that sits in NEW, not here.) Habit-trap: posting a full portfolio as a marketing necessity now doubles as handing over training data, and pricing/contract norms still assume that showing work doesn't give it away.

Unchanged axioms

1. A named illustrator is who a client trusts to interpret a brief, take direction, and stand behind the result. A LoRA reproduces the look but can't be answerable — can't be briefed on what the client actually meant, can't be told "this reads wrong for the audience" and course-correct, can't carry the liability when the image ships and is off. The trusted, named, accountable person is what the style-clone conspicuously isn't. This gets more load-bearing as the images themselves commoditize.
2. Iterating live on a brief is part of what's bought. The scarce act was never the rendering; it was reading an underspecified ask, pushing back on it, and converging on what the client couldn't articulate. A model generates a hundred variants but doesn't own the judgment of which one is right for this client, this context — and can't be the party in the room negotiating the intent. Calibrate: agentic models are getting better at multi-turn revision, so the narrow claim is that taking accountable direction on an ambiguous brief stays scarce, not that "generating revisions" does.
3. Legal and consent standing over one's own style is something only the artist can hold. Whatever regime settles — licensing, opt-out, style-right — the party who can grant or withhold consent, sign the release, and sue over an unlicensed clone is the human author, not the model and not whoever ran the training. Standing to make and enforce that commitment is non-transferable. This is fast-moving: it hinges on unsettled law and platform policy in 2026, so it's standing-in-principle, not a reliable enforced right yet.
4. A recognizable style, held as the identity of a trusted maker, is still an asset — but the asset is the authorship, not the look. The look is now copyable; what a client can't clone is the ongoing relationship with the person whose name the style belongs to, who keeps developing it and answers for it. The moat moved from the style being hard to imitate to the author being the only legitimate source of it.

New axioms

1. Style-cloning without consent. A public portfolio is now training data, and a client or third party can generate "more in your style" without commissioning — or paying — the artist at all. There's no settled mechanism to consent, license, decline, or get paid when your own back catalogue is used to compete with you. This is the fury in the market right now, and it has no working answer.
2. Provenance and licensing of a personal style. When output in an artist's style is abundant and detached from them, the scarce, valuable thing becomes a credible "this was authored/authorized by them" — and there's no established way to attach that, or to license a style as a controlled asset the way a font or a brand mark is licensed. Whoever builds trustworthy style-provenance defines the terms of the next market.
3. The commodity-illustration commission tier collapses, and the ladder that ran through it breaks. The tier that many illustrators lived on — and that early-career artists used to build a book, a reputation, and eventually a name — is being undercut to near-zero. If the entry rungs disappear, it's unresolved how a new illustrator accumulates the reputation and trusted-authorship that STILL HOLDS rewards. The scarce thing survives; the path people used to reach it may not.

Where it breaks

The client arrives with an AI mockup in the illustrator's own style and asks for that, at that price and that speed (INVALID: the image-as-deliverable was the sale) — while what they actually still need, and the only thing they can't get from the mockup, is an accountable author who'll take the brief and stand behind the result (STILL HOLDS: authorship and live iteration). The transaction is being negotiated over the commodity that collapsed, so the illustrator is pushed to compete on the one axis they've already lost instead of the one they still own.

Second collision: the portfolio an illustrator must post to win the trusted-author work (STILL HOLDS: named authorship is the moat) is the exact dataset that trains the clone which undercuts them (NEW: consentless style-cloning). Marketing the authorship and surrendering the style are currently the same act, and nothing separates them yet.

ai-axioms.vercel.app/axioms/media/illustration-commission-ai-axioms