Ethics

Compose vs generate: two very different kinds of AI

31 Jul 2026 · 8 min read

Generating AI synthesises new imagery from training data; composing AI arranges work that already exists. The difference is authorship: a generator invents pixels with murky provenance, while a composer places a creator’s finished, owned artwork. That gap decides copyright, credit, and whether marketplaces will permit it.

“AI” has become a single word doing two opposite jobs. One kind of system invents imagery that did not exist before. Another arranges imagery that already does. We call these generating and composing, and conflating them is the source of most of the confusion — and most of the ethical and legal risk — in creative commerce right now.

Defining the distinction precisely

A generative model produces new pixels. You prompt it, and it synthesises an image from patterns absorbed across a vast training corpus. Nothing in the output existed before; its provenance is statistical, distributed across everything the model ever saw.

A compositional system does not invent imagery at all. It takes specific, finished inputs — a creator’s existing artwork — and performs operations on them: place, scale, crop, mask, tile, mock up onto a product. Every pixel of the artwork traces back to a known author. The AI’s job is arrangement and admin, not creation.

Generate: synthesises new imagery; provenance is diffuse and contested; authorship is murky.
Compose: arranges existing, owned imagery; provenance is exact; authorship is unchanged.
Generate answers “what should this look like?”; compose answers “where does this finished art go?”
A generator can produce work in someone’s style; a composer can only place that someone’s actual work.

Why the distinction is not pedantry

It is tempting to wave this away as semantics. It is not. The two approaches sit on opposite sides of nearly every question that matters in creative commerce: who holds copyright, who gets credited, who gets paid, and whether the work is allowed on a platform at all. Get the category wrong and you inherit the wrong answer to all four.

What it means for artists’ rights

When a model generates, the relationship between output and any individual creator is contested — which is exactly why the courts are busy. Recent years brought a wave of high-stakes copyright litigation over training data, including very large settlements and suits alleging models were built on pirated work. The unresolved question across these cases is whether generated output substitutes for the works it learned from.

Composition sidesteps that entire fight by construction. There is no training-data ambiguity because nothing is learned and nothing is synthesised. The artwork is the artist’s, before and after. Copyright stays put. Credit stays attached. Income stays with the author. The clean answer is available precisely because the system never tries to create.

This is Realform’s whole architecture: compose, never generate. We never synthesise imagery in a creator’s style, or near it. Agents place a creator’s existing, finished artwork onto made-to-order products and run the surrounding business — so authorship, copyright, credit, and income never leave the human.

Why it matters for marketplace compliance

Marketplaces have drawn the same line. Etsy’s standards require sellers to disclose generative-AI use, select “Designed by” rather than “Made by,” and tick an AI-generative checkbox — with undisclosed listings flagged and removed by automated moderation. The rules treat generation as a category that must be declared and constrained.

Composition of a seller’s own original artwork does not trip those wires, because there is no generated content to disclose and no third-party authorship to untangle. Building on composition rather than generation is therefore not just an ethical preference — it is the lower-risk path through a compliance landscape that is tightening every quarter.

Choosing the right kind of AI

If your value is your authorship, generation is the wrong tool to aim at the art: it dilutes the very provenance you are selling. Composition keeps that provenance intact and points the machine at everything around the art instead. The question to ask of any “AI-powered” creative tool is simple — does this invent imagery, or does it arrange mine? The answer tells you who will own the result.

FAQ

What is the difference between composing AI and generating AI?

Generating AI synthesises brand-new imagery from training data, so its provenance is diffuse and authorship is contested. Composing AI takes a creator’s existing, finished artwork and only arranges it — placing, scaling, mocking it onto products. Nothing new is invented, so authorship stays exactly where it started.

Why does compose versus generate matter for copyright?

Generation raises unresolved questions about training data and whether output substitutes for the works behind it — the core of ongoing litigation. Composition avoids that entirely: the artwork is the creator’s before and after, so copyright never moves and there is no training-data ambiguity to fight over.

Do marketplaces treat composed and generated work differently?

Yes. Platforms like Etsy require disclosure of generative-AI use, a “Designed by” label, and an AI checkbox, with undisclosed listings removed automatically. Composition of a seller’s own original artwork has no generated content to disclose and no third-party authorship, so it sits on the lower-risk side of those rules.

Can a composing system create art in my style?

No, and that is the point. A composer can only place your actual, finished work — it has no ability to synthesise imagery in your style or anything resembling it. That constraint is what keeps your authorship, credit, and income intact rather than diluting them.