The First AI Generated Image: The Lesser-Known Story

Before we dive into the history of AI images, we should define what they are. The thing is, there are two definitions for this term, one more limiting than the other.

The modern definition of an AI image posits that it’s an image produced by artificial intelligence models based on the user’s input (prompt). The models in question rely on vast databases of existing images to produce new, never-before-seen ones.

Historically, however, the term “AI image” could refer to images generated with rule-based AI programs that relied on hand-coded capabilities. In other words, modern AI image generators learn on their own from databases, but this wasn’t the case for AI programs in the 20th century.

Contemporary generative AI models, such as generative adversarial networks (GANs), use machine learning algorithms to find patterns in vast databases of labeled images. For example, to “learn” what a dog looks like, the model analyzes thousands of dog images and identifies common pixel patterns.

To generate an AI image, you can either enter a text prompt to generate an image or upload an existing image as an example. The model recreates the pixel patterns to present you with original visuals.

Curious about the history of common misconceptions? Check out our blog post on what is homework backwards!

With this brief refresher done, let’s move on to our headliner question: What is the first AI generated image? The consensus is that this title belongs to artworks generated by AARON, several computer programs created by Harold Cohen in the 1960s and 1970s.

Cohen, a painter himself, started developing AARON using a symbolic AI approach. The idea was to create a program that would autonomously generate original artworks in his style, which was hand-coded into the program. The first AARON exhibition in 1972 consisted only of black-and-white works because the program couldn’t work with color yet.

At around the same time as Harold Cohen started his work on AARON, Frieder Nake began his journey into computer-generated art, as well. In 1963, as a University of Stuttgart student, he was asked to develop software for a new drawing machine set to arrive at his university.

That dallying with drawing machine software sparked his interest in what is now known as algorithmic art. This type of art relies on algorithms that use complex calculations to generate patterns that constitute the work. (Fractal art is an example of algorithmic art.)

Frieder Nake’s programs used points, lines, and geometric shapes to create works. The programs themselves included:

• compArt ER56 (1963-65)
• Walk-through-raster (1966)
• Matrix multiplication (1967/68)
• Generative aesthetics I (1968/69)

Another pioneer in computer-generated art, Georg Nees wrote his PhD thesis on generative art and design. Trained as a mathematician, Nees pivoted into philosophy after a career at Siemens.

Published in 1969, the thesis, titled Generative Computergraphik, included:

• Examples of computer-generated graphics
• Snippets of the code that generated them

That said, even before the thesis was finished, Nees exhibited some of the world’s first computer graphics as works of art in 1965. These graphics can also compete for the title of the first AI-generated image if you expand the definition to include rule-based systems infused with randomness-generating algorithms. After all, Nees's geometrical patterns were based on random numbers generated by a rule-based program.

Together with Nees and Nake, A. Michael Noll is known as one of the “3N” pioneers in computer graphics. An engineer by education, Noll worked in a variety of trades, from marketing to basic research. But he left his arguably greatest mark on digital computer art.

As a researcher at Bell Labs, Noll blazed a trail for computer-generated 3D graphics and animation. Together with Noll and Nees, he exhibited his computer art, generated by the code he wrote, in New York in 1965.

Of course, that code was far from the sophistication of today’s geometry AI solver or image generator. Still, for his time, Noll’s computer-generated animated title sequences and artworks were revolutionary.

Chances are, the first AI generated image ever is lost to time, forever. It’s because the true first was probably created in lab conditions during a trial run and discarded by its creator as they were refining the program. So, all we can find is the first AI-generated image that was made public, which is an important distinction.

The present of AI images reminds us how far the technology has come. DALL-E alone was reported to generate two million images a day in 2022, but the tool’s popularity has grown since. One 2024 investigation estimated that, across tools, users generate 34 million images every single day. In 2026, AI images are more realistic and sophisticated than ever, and they are everywhere.

Trying to predict what’s next in tech is usually a thankless job. That said, advances in generative AI models and computing power are steadily paving the way for real-time video generation, interactive editing, and AI-generated assets use in AR/VR. But for now, the debate around intellectual property rights, potential for misuse, and environmental impact continues.

Only time can tell how generative AI will evolve in the years to come.