What is AI art? An honest introduction

AI art is art in which artificial intelligence is not merely a tool but part of the artistic procedure itself: a system that generates images, sounds or movement from probabilities, steered by a human who decides what survives. The boundary is blurry, and that is not a flaw: photography, too, was once "not art", until it became art.

How generative images are made

Most systems behind today's "AI art" are diffusion models. They learn from billions of images with text captions which noise belongs to which concept, and then generate new images by resolving noise into form, step by step. The prompt is not a magic formula but a search through a learned space: same model, same sentence, four different results. The artistic decision lies in the selection, the iteration and the context, not in the button press.

Neural networks, GANs and the rest of the family

Before diffusion models, GANs, generative adversarial networks, dominated: two networks competing, one generating, one judging. The famous "Portrait of Edmond de Belamy" auctions by Obvious (2018) and Mario Klingemann's surreal faces were made this way. Alongside runs the much older tradition of generative art: rule systems, chance and procedures, from Sol LeWitt to Vera Molnár, who became an icon of algorithmic art at 96, shortly before her death in 2022. AI art is that tradition's most radical continuation, not its opposite.

Who is the author?

The most uncomfortable question. The practical answer: the author is whoever shapes the procedure. A prompt alone is rarely art, just as a single brushstroke rarely is. Curators and courts are slowly converging on a distinction between tool use (human authorship remains), co-creation (shared authorship, legally mostly unsettled) and autonomous generation (most likely no copyright at all). For the visitor the more interesting question is different: has someone found a form in which the machine makes visible what would otherwise stay invisible?

What makes a good AI exhibition

There are three degrees of honesty. In the first, AI is used as an effects machine, pretty images that could be a screensaver loop. In the second, the model becomes the medium: its noise, its errors, its biases are made visible. In the third, the system itself is the subject: what does it mean that a machine learns what a face is? The exhibitions listed in this directory span all three degrees, which is legitimate; you should just know which one you are looking at.

Signs of strong programmes

Works generated live instead of finished videos, processes that happen in the room and never repeat. Curated datasets with named sources. Artists who built or trained the system rather than merely using it. And language that explains what the model does without mystifying it.

Where to see AI art live

The scene concentrates in a few cities. In Berlin, the AuraLabs run a small permanent exhibition on AI and art, and the P61 Gallery presents digital art at canvas scale. In Hamburg, from October 2026 the major Hito Steyerl survey at the Deichtorhallen honours the sharpest analyst of the politics of the digital image. In Madrid, the Fundación Telefónica makes the topic free of charge. Our city guides for Hamburg, Berlin and Madrid collect all the dates.

Why this is not a hype (and why it is)

The market for AI art had its first echo in 2022 and has since calmed down, as happens with every new medium: first scandal, then routine, then self-evidence. What remains is the question of machine images, and that question is not getting smaller. Anyone walking into a good AI exhibition today is not watching future music but the present learning how it sees itself.

A selection of the season's key dates is in our autumn 2026 roundup.