What Is Generative Art?

Generative art is art made by a system the artist designs: the artist writes the rules, and the rules make the image. Chance is usually built in, so the same program produces work that is never identical twice and never outside the boundaries its author set. The Tate defines it as art made using a predetermined system that often includes an element of chance, usually applied to computer-based art.

It is one of the oldest forms of computer art, with a documented practice running back to the 1960s, and one of the most frequently misdescribed, because it gets collapsed into AI image generation, which works in an entirely different way.

What is generative art?

Generative art is art produced by an autonomous system that the artist built. The artist defines the process; the process produces the artwork. That division is the whole discipline.

A generative work has two parts: the system, and what the system outputs. A painter decides where every mark goes. A generative artist decides what kinds of mark are possible, how marks relate to one another, where chance is allowed in and where it is held out, and then accepts what resolves. The authorship sits in the rules rather than in the surface.

This is why the same piece of code can yield a thousand distinct outputs that are all recognisably one artist's work. The variation is intended. What the artist controls is the space of possible images, not any single one of them.

How is generative art different from AI art?

Generative art runs on rules the artist wrote. AI image generation runs on a model trained on other people's images. In the first, every decision in the system can be accounted for by the person who put it there. In the second, the decisive work was done by a training process on material the artist did not make and often did not licence.

The outputs can resemble each other. The authorship does not. A generative artist can hand you their source code and it explains the work completely, including why it can do what it does and nothing else. That is not available for a prompt.

The distinction matters commercially as well as ethically. It is why generative art has a collecting history stretching back six decades while text-to-image tools are a few years old, and why institutions including the Tate, the Whitney Museum of American Art and the Centre Pompidou hold generative work in their collections.

Where did generative art come from?

Generative art began with plotters and mainframes in the 1960s, decades before personal computers. Harold Cohen, whom the Tate names among the first practitioners, used computer-controlled robots to generate paintings in the late 1960s.

Vera Molnar started making rule-based drawings in the same decade, first by hand using a method she called her machine imaginaire, then with a computer once she had access to one. Frieder Nake, Georg Nees and Manfred Mohr were working along the same lines. What that generation established is still the shape of the practice: the artist writes a procedure, and the procedure draws.

Molnar's standing was confirmed late in her life. In 2023, at the age of 99, she released Themes and Variations, a series of 500 generative works made with the artist and creative coder Martin Grasser. Sotheby's sold the series through a Dutch auction, the first in the house's 300-year history, and it sold out in under an hour for 631 ETH, around 1.2 million dollars.

What tools do generative artists use?

Most generative artists write code, and much of the software they use was built by artists for artists rather than adapted from engineering tools.

In 2001 Casey Reas and Ben Fry released Processing, an open-source language designed for visual work, and it brought a generation into the field. Reas founded the Processing Foundation in 2012, has been a professor in UCLA's Department of Design Media Arts since 2003, and his work is held by the Whitney Museum of American Art, LACMA, SFMOMA, the Centre Pompidou and the V&A. Zach Lieberman co-created openFrameworks toward the same end, co-founded the School for Poetic Computation, and leads the Future Sketches group at the MIT Media Lab.

That habit of building the instrument continues. Karsten Schmidt has written open-source libraries used across the field and was an early Processing contributor. Josef Pelz built T3D, which made 3D texture work available to people without a graphics programming background. Anna Lucia, an engineer by training, writes software that drives an embroidery machine, translating quilting into algorithm. TouchDesigner, p5.js and openFrameworks are the common environments.

What does generative art look like today?

Contemporary generative work is usually time-based. It is not a finished picture but a process that keeps running, so what is on screen in the morning differs from what is there at midnight.

That creates a practical problem. A still image of a generative work is a photograph of it. A video file is a recording of one pass through a system that was built to continue. Neither is the work. This is the single most common way generative art is misrepresented, and it is why the field has pushed toward displays that can execute code rather than replay footage.

Artists working this way now include Leander Herzog, making real-time animation and web-based work since 2006; Per Kristian Stoveland, who co-founded the Oslo studio Void and released The Harvest through Art Blocks; JERES, who came to the field from computer engineering and art history; and Jason Ting, whose generative animation draws on James Turrell, Dan Flavin and Vera Molnar.

What is the difference between generative, algorithmic and procedural art?

The three terms overlap and are often used interchangeably, but they emphasise different things. Generative art is the broadest: any art produced by a system with some autonomy, whether that system is code, a set of written instructions, or a physical process.

Algorithmic art narrows to work where the system is an explicit algorithm, usually mathematical. Procedural art, a term more common in games and film, describes content produced by a procedure at the moment it is needed rather than authored in advance. A procedurally generated landscape and an algorithmic drawing are both generative; not every generative work is algorithmic in the strict sense.

Sol LeWitt is the useful edge case. His wall drawings were sets of written instructions executed by other people, with no computer involved at any point. They are generative by the definition above, which is a reminder that the system matters more than the silicon.

How do you start making generative art?

Most people start with Processing or p5.js, the JavaScript successor, because both were built for visual work rather than adapted to it and both are free and open source.

Processing runs locally and suits print and high-resolution output. p5.js runs in a browser, which makes it the shorter path from an idea to something shareable. openFrameworks, written in C++, is the choice when performance matters, typically for installations and live performance. TouchDesigner takes a different approach, building work as a visual node graph rather than as written code, and has become the common environment for real-time and installation work.

The technical barrier is lower than most people assume, and it is not where the difficulty sits. Drawing a thousand random lines takes a few minutes to learn. Deciding which thousand lines are worth looking at is the practice, and that part takes the same years any medium takes.

What is long-form generative art?

Long-form generative art means a system released as a fixed edition, where the artist publishes the algorithm and each collector receives a distinct output from it that neither party saw in advance.

The term came out of Art Blocks, the platform that popularised the format. It raises the stakes considerably: the artist has to be satisfied with every output their system can produce, not just the ones they would have chosen, because the full range goes out under their name. A system that produces ten excellent images and ninety poor ones fails as long-form work.

Sotheby's runs its Gen Art programme in partnership with Art Blocks specifically around this format, committing to exclusive long-form sales from two to three artists a year. It is the clearest sign that the discipline of designing a whole possibility space, rather than curating outputs after the fact, is now what the field is judged on.

Is generative art valuable?

Generative art has an established collecting market at both institutional and auction level. Museums including the Whitney, MoMA, the Victoria and Albert Museum, SFMOMA and the Centre Pompidou hold generative and computational work.

At auction, Sotheby's runs a dedicated Gen Art programme in partnership with Art Blocks, committing to exclusive sales of long-form generative work from two to three artists a year. The Molnar sale in 2023 realised approximately 1.2 million dollars for 500 works in under an hour.

Value in this field tracks the same things it does elsewhere: the artist's record, the significance of the system they built, and whether the work can be properly experienced. The last of those is specific to the medium, and it is a reason display quality matters more here than for a print.

How do you display generative art properly?

A generative work needs a computer to exist. If it is rendering live, something has to run the code, and the specification of that hardware decides what you see: colour depth, black level, resolution, and whether the motion holds together without stutter.

Most screens in a home are televisions. A television is built to decode and replay a video signal, which is the wrong job. Playing a generative work as a video file on a television reduces a running system to a loop, and the loop is visible: the same sequence returns, and the sense of a work that is thinking stops.

The Layer Canvas was built for this. It carries a 3.2 GHz CPU and GPU and renders generative work in real time on a museum-grade square panel, so the art runs rather than loops. It has no tuner, no apps and no camera. See Layer Curation for the collection, or what living art means for why a running work is a different object from a recorded one.

Can you own generative art?

Yes, and ownership in this field works differently from owning a print. What a collector acquires is usually the right to run a specific output of a system, together with the provenance that ties that output to the artist who designed it.

Because the work is a process rather than an object, the questions a collector asks are different. Does the piece render live, or is what I have a recording? If it renders, what hardware does it need, and what happens when that hardware is obsolete? Who holds the source, and is it escrowed anywhere? Museums acquiring computational work have been asking these questions for years and have developed conservation practice around them.

This is the part of the field that is least settled and the most interesting. A painting's conservation problem is physical decay. A generative work's conservation problem is that the platform it was written for stops existing, which is why open-source tooling and published source matter to institutions as much as to artists.

Why does generative art suit a dedicated display?

Because the work has no fixed final frame, the display is not showing a reproduction of the art. It is the place the art happens.

That makes the specification part of the viewing experience in a way it is not for a photographic print. Black level decides whether a dark composition reads as depth or as grey haze. Colour gamut decides whether the palette the artist chose survives. Refresh behaviour decides whether slow movement looks continuous or stepped. Processing power decides whether a complex system can run at full resolution without dropping detail to keep up.

A square format matters more here than elsewhere too. Generative work is frequently composed to a square because the form has no cinematic convention behind it, and cropping it to a widescreen panel discards part of the composition or adds bars around it.

Who are the best generative artists to follow?

The field has a canon and a working present. From its history: Vera Molnar, Harold Cohen, Frieder Nake, Georg Nees and Manfred Mohr.

Working now: Casey Reas, Zach Lieberman, Karsten Schmidt, Anna Lucia, Leander Herzog, Per Kristian Stoveland, Josef Pelz, Newnome Beauton and JERES. A fuller list of artists, with biographies and work, is on the Layer artists page.

Sources

Tate, Art Terms: Generative Art, tate.org.uk/art/art-terms/g/generative-art. UCLA Department of Design Media Arts, faculty profile: Casey Reas, dma.ucla.edu/people/casey-reas. ARTnews, Vera Molnar NFT Auction at Sotheby's Realizes $1.2 M. Total, 2023. Sotheby's Gen Art programme with Art Blocks.