A gallery hangs a large-scale, algorithmically generated visual piece next to a hand-painted canvas, and visitors rarely agree on which one required more genuine creative decision-making. That disagreement is not a failure of the audience to understand the medium. It is a fair reflection of an unresolved question the art world has not settled either: when an algorithm generates the image, where exactly did the creative act happen, in the code, in the prompt, or in the choice of which output to keep?
It is worth separating this question clearly from a related but distinct one. Generative design, the use of algorithms to explore engineering solutions for weight, strength, or manufacturability, which we cover in our piece on generative design and algorithmic engineering, is judged against measurable, functional criteria: does the part hold the load, does it fit the machine. Generative art has no such objective test. It is judged against taste, meaning, and context, categories an algorithm has no native access to.
Where the creative decision actually lives in a generative artwork
In a purely algorithmic piece, artists such as those working with generative code have long made the case that the creative act lives in designing the system’s rules, the parameters, the constraints, the aesthetic boundaries within which the algorithm is allowed to explore, rather than in any single output the system happens to produce. That framing predates today’s AI image generators by decades, going back to early computer art and generative music pioneers who wrote the rules and then let a system run within them.
What changed with large-scale AI image generators specifically
Tools like Midjourney compress that rule-writing process down to a short text prompt, which raises the automation question much more sharply than earlier generative art did. Writing a few descriptive words is a far smaller creative contribution than authoring an entire generative system from scratch, which is exactly why critics distinguish between “prompting” and “authoring” when they argue AI image generation leans closer to automation than to art.
Entering prompts, no matter how detailed, is not enough to establish the kind of creative control that copyright law, or most working artists, recognize as authorship.
What the law has actually decided so far
This is not purely a philosophical debate anymore. In March 2026, the US Supreme Court declined to hear a further appeal in Thaler v. Perlmutter, leaving in place a lower court ruling that a purely AI-generated image, with no human creative input beyond initiating the process, cannot be copyrighted under US law. The US Copyright Office has clarified separately that AI-assisted works can qualify for copyright protection, but only where a human exercises meaningful creative control over the expressive choices in the final piece, not merely by entering a prompt and accepting the first result.
Two working models of generative art, compared
| Approach | Where the human decision sits | Closer to |
|---|---|---|
| Custom generative code (rules-based) | Designing the system’s constraints and logic | Authorship |
| Single text-to-image prompt, first result kept | Selecting from an existing model’s output space | Automation |
| Iterative prompting, editing, and compositing | Curating, refining, and combining outputs deliberately | Somewhere between the two |
Why the middle category is where most practicing artists actually sit
Very few artists working with these tools today fit neatly into either extreme. Most use AI generators as one stage in a longer process: generating raw material, then curating, editing, compositing, and recontextualizing it into something the model alone would never have produced unprompted. That layered process is closer to how a photographer works with a camera, a tool that automates image capture but does not automate the decisions about framing, timing, and selection that make a photograph worth looking at.
What this means for how the work should actually be judged
Judging a piece of generative art by asking only “did AI make this” skips the more useful question: how much deliberate human decision-making shaped the final result, and is that decision-making visible in the work itself. A collector, a gallery, or a curious viewer is better served asking about process than about tool, since the same tool can sit anywhere on the spectrum from pure automation to genuine authorship depending entirely on how it was used. The same process-over-tool question is what we work through from the applied, commercial side in our piece on AI versus human creativity in graphic design, where the stakes are less about authorship in the abstract and more about what a client is actually paying for.
Our take: generative art is neither pure automation nor pure creativity by default, it becomes whichever one the artist’s actual process earns. The tools will keep improving, the legal questions will keep evolving case by case, but the underlying test has stayed remarkably stable so far: did a person make deliberate, expressive choices that shaped this specific result, or did they simply accept whatever the system handed back first.
A question worth asking before dismissing a piece either way
Before writing off a generative piece as automated, or defending it as pure authorship, it helps to ask a concrete question: could the exact same result have come from a different artist typing a similar prompt into the same model. If the answer is genuinely yes, the piece leans toward automation regardless of the label attached to it. If the specific sequence of curation, editing, and recombination the artist applied would have produced a meaningfully different result in someone else’s hands, that is a stronger signal of authorship than the tool used to get there.