What is neuro art? Will the neural network be able to replace the artist?

In the interactive exhibition "Alice in the Neuro Kingdom", the creators offered to look at the works of the popular English writer Lewis Carroll through the prism of the Russian visual tradition. This is a multi—genre mix of illustrations, historical artifacts and traces of Alice in the history of Russia - from Vladimir Nabokov to Vladimir Vysotsky. Multimedia design and video cards were created using neural networks that processed images of Lewis Carroll in the style of Ivan Bilibin, a famous illustrator of Russian fairy tales. We talked with the artist Valeria Titova about how neural networks can already draw an image from text using a schematic sketch or a sketch.

— To begin with, what is neuroart?

— This is any artistic statement created with the help of neural networks.

— Then how do you work with a neural network?

— Usually I go from an idea or from the progress of work. I am either interested in looking for technical possibilities of embodying my thoughts — and then I try a familiar program, write scripts from text prompts for a neural network (promts), select references (initial sketches, sketches) for the visual part, and then implement. Either it's a new technology — and I get something out of my ideas or permanent motives of creativity.

— So AI is an artist's tool? Or his co-author and even the author of the work?

— Of course, this is only a tool in relation to the artist. Just as the co-author is not a painter's brush and a photographer's camera.

— Then how does a neuro-artist differ from someone who uses similar tools and downloads, for example, the Lensa application, and then creates new images from their photos?

— I believe that everyone is an artist. If he feels a creative impulse, thinks and draws conclusions while processing photos in Prisma, Lensa or somewhere else, let him be an artist. We are talking about the self-identification of a person as a creator and his choice to create even by pressing one button.

What is Lensa?
Lensa is a mobile application developed by the Russian company Prisma Labs. It appeared about a year ago and was originally a standard photo editor. In November 2022, an updated version was released, which added a feature that makes people in photos look like movie characters, cartoons and comics. By uploading a dozen of your photos to Lensa, you can get from 50 to 200 generated portraits.

— Images are born from text. It turns out that formulations play an important role. Does the artist become a writer in this case?

— Yes, words play a special role, but this is not a literary language at all, but the language of a text query. I think an artist is a broad concept that can accommodate all the details of the process — external and hidden. For the last century, artists have been working not only with visual mediums.

— What is the role of unpredictability in working with neural networks?

— I don't work very carefully and by inspiration, so if we talk about my work, there are a lot of random finds there. At the same time, I understand how artistic styles, light and materials can be displayed — there are special tables and libraries for this, where all this texture is collected.

— Images created by neural networks are sold at auctions on a par with paintings painted by people. How do you feel about this? Will the neural network be able to replace the artist?

— Auctions are a separate topic, where other contexts and special rules are layered on art. I feel good about it — this is how neural networks are legitimized, and developers receive more funding. It's great for our common future.

By itself, the neural network will not be able to replace anyone. For a good, not even excellent, but just a good result, a creative erudite person should sit behind it. Otherwise, the work will not go well.

— At what level can you really control the aesthetics of the generated image?

— From the point of view of images, this can be controlled. It all depends on the erudition or patience of the author. It is enough to know artists and techniques, to understand how the light works and the camera works, to write good scripts, or to look for examples of what an idea might look like in libraries. For high-quality control over the work and its visual style, you need to have the desired image, and it can only come from the artist's head.

— The neural network works only with a set of already existing images. It turns out that she is not able to come up with anything fundamentally new?

— There is nothing fundamentally new. To do this, an artist is sitting behind it.

— The aesthetics of neuro-art are often psychedelic. What could this be related to?

— I think it's because the neural network turns some objects into others, and a person is used to seeing this in a state of altered consciousness or in a dream. At the same time, the technology is good enough that many items are recognizable and look like real ones. Soon, videos created by neural networks will become indistinguishable from those captured on camera and will not have their own specific aesthetics.

— The technical side of neuro-art is developing very quickly. How do you see his immediate future?

— If humanity once again does not get scared and does not slam on the brakes and the "winter of AI" does not come, then we are waiting for a lot of understandable tools for any level. I myself want to try the text2video tools to create not just frame-by-frame animation, but full-fledged films and cartoons with a plot. I'm waiting for all this to be introduced into every production along with simpler neural tools that everyone is already used to.

Five AI brushes for a novice neuro-artist

Valeria Titova has compiled a list of programs for "S (W", which she uses herself and recommends to beginners:

Phygital Plus. A convenient tool is a simple interface, a lot of neural networks "under the hood", constant updates, as well as a neural network library that will make it clear what neural networks can do today.

Midjourney. The most aesthetic neural network for generating images with the "beautiful" button.

StableDiffusion. The easiest way to play with the text-to-image conversion model.

Runway. A tool with many convenient functions for video processing.

DALL-e 2. This neural network for generating images, although it appeared more than two years ago, does not lose relevance today, and it is worth trying.