Images created with artificial intelligence have proliferated and improved in recent years, producing wildly detailed and imaginative results. Which may have made you wonder, “How does AI art work, anyway?”
AI art is, simply, artwork made with the assistance of generative AI — a technology that finds patterns in big datasets and uses that information to create new content. All it takes is an AI art generator, like Adobe Firefly and an idea. The artist types a detailed prompt, which the tool then uses to generate image options based on the description.
Prompt: Three eiffel towers in desert with river; photorealism
There is more than one kind of generative artificial intelligence.
For writing anything from poetry to an email, there are large language models, which are trained on text and help people generate copy. To create illustrations, painterly creations, logos and more, diffusion models trained on images help people make artwork of all kinds. Many of the popular generative AI tools use one of these models.
Artists use generative AI to make a variety of different artwork, from poems and stories to creations that look like analogue paintings or photographs and more. The speed and flexibility of generative AI enables creators to jumpstart and finish projects more quickly and opens up all kinds of exciting new avenues for creative expression.
How does AI generate art?
People create art from the things that surround them — trees in the forest, cityscapes, their own reflection in a mirror. Generative artificial intelligence also takes in lots of information in the form of words and images and uses those to create artwork from a prompt.
The technology that powers this ability is called a neural network. A neural network is a mathematical system — an algorithm — that finds patterns in big sets of data. When you prompt an AI generator to depict a tree, it’s using the information it has learnt about what trees look like to create a new image. And it’s your guidance as the artist that refines those images even further - telling it to generate a pink fir tree, for instance or a tree blooming with tropical flowers. These tools are packed with information, but it takes the imagination of the user to create artificial intelligence art.
Prompt: (left) pink fir tree; (right) tree blooming with tropical flowers
Generative AI in the art world.
Artists have been experimenting with forms of artificial intelligence in their work for decades, pondering the implications and applications of such technology long before the general public was.
Artist Vera Molnár began experimenting with early programming languages to produce randomly generated artwork in 1968. Considered a pioneer in generative art, her geometric creations are included in major museum collections.
Flash forward a few decades and you can find AI generated art within the Museum of Modern Art in New York. In 2023, the museum displayed the exuberant, ever-changing artwork of Refik Anadol on an enormous screen. The abstract creation was made with artificial intelligence trained on artworks held within the museum’s collection.
AI artworks have been auctioned by Sotheby’s and exhibited at the Venice Biennale. Artificial intelligence has been folded into arts course curriculum at institutions like Columbia University and the Rhode Island School of Design.
Prompt: a beautiful gallery oil painting, red, jade, orange and grey, sharp lines and blended tones
Get to know the technology behind AI-generated art.
Just like there are many tools in an artist’s physical toolbox — paints, pencils, inks — mulitple kinds of technology contribute to making AI artwork. We’ve already discussed large language models and diffusion models. Here are a few more technologies used in AI art generation.
Generative AI models
Generative adversarial networks (GANs): GANs are a set of two neural networks, trained on the same data, that work together. One generates a photorealistic image and the other tries to figure out if that image is real or created. For instance, the first may generate an image of a horse and the second would try to determine if the image was a photograph or digitally created. This helps the system make more realistic images.
Variational autoencoders (VAEs): A variational autoencoder is made up of two neural networks working in tandem, each with a different job. One is an encoder, which takes in information and the other is a decoder, which is able to reinterpret that information into all new content. Like GANs, they produce photorealistic images.
Generative AI techniques
Image synthesis: Image synthesis describes the act of creating new images from large datasets of other images.
Creative coding: Artists who create the programmes that generate their artwork are described as creative coders. The intent of creative coding is not to produce a functional result, but an expressive one.
Join the generative AI art revolution.
Now it’s time to put all your knowledge to good use and explore what’s possible when you combine the power of imagination with the abilities of generative AI.
Experiment with new artistic styles.
With generative AI it’s easy to get outside your artistic comfort zone. Explore new looks with Firefly’s Text to Image generator by choosing options from the Style menu like Steampunk, Layered paper or Stippling.
Save time, create more.
Use generative AI to quickly create mockups for clients or reference images for your artwork. Use Firefly powered tools like Generative Recolor and Text to Pattern (beta) within Illustrator to quickly iterate on your art and create stylish motifs.
Share work with others.
Invite a friend to edit one of your artworks using Generative Fill in Photoshop, a Firefly powered tool that lets you select any part of an image and replace it using a simple text prompt. With generative AI, you can create with others like never before.
Creativity for all.
Generative AI gives you access to more types of artistic expression. A skilled graphic designer using AI can try their hand at making dramatic edits to a photograph or a total beginner can experiment with multiple kinds of artmaking using Text to Image.
Get started with creating AI artwork.
1 Open the Firefly web app.
The simplest way to begin experimenting with generative AI is to use Adobe Firefly. Using the Firefly web application you can immediately begin creating artwork, recolor your existing artwork or edit photographs using generative AI. To get started go to Firefly.adobe.com and sign in. If you don’t have an Adobe ID, you can create one for free here. You can also explore features powered by Firefly like Generative Fill in Photoshop, Generative Recolor or Text to Pattern (beta) in Illustrator and Text to Image and Text Effects in Adobe Express.
2 Write a prompt.
Writing a good descriptive prompt is one of the key components of creating generative AI art. Use adjectives, include the names of art movements or mediums and try including moods and emotions as well. Learn more about writing successful AI art prompts.
3 Explore sample prompts.
Mouseover the Text to Image artwork examples provided in the Firefly web app gallery to see what prompts produced those images and apply those techniques to your own experiments.
4 Refine your settings.
Explore the options available within the Firefly web app to refine your output. You can choose from a selection of different Content Types, Styles, Colour and Tone, Lighting and Composition options.
5 Keep experimenting.
Once you’ve generated an image, see how much further you can take it. The creation of an artwork with generative AI can be a catalyst for exploration. Use it to inspire an analogue work or take it into another Adobe app for fine-tuning with precise editing tools. Learn more by following our step-by-step guide on how to make AI-generated art.
Unlock your creativity with Adobe Firefly and AI-generated art.
Artists have been experimenting with artificial intelligence for decades, exploring how it can change the way we make and view art — and now this technology is readily available to all. Get started with Firefly today and see where your imagination takes you.
Questions? We have answers.
The current Firefly generative AI model is trained on a dataset of Adobe Stock along with openly licensed work and public domain content where copyright has expired.