Does a neural network draw better than artists?

Neural networks gradually learned how to draw cool art and process photos. Midjourney, NightCafe, Dall.E-2 and other AI for creating images on the prompts of the user generate cool images and soon artists will not be needed. Let's check if this is true?

What is a neural network and how does it work

Recently, all you can hear is that artists are no longer needed and will be replaced by neural networks. As soon as they appeared, neural networks generated only bizarre pictures that could not be called art. However, now there are many AIs that can generate a full-fledged high-quality image in a few seconds.

How does a neural network create an image? Very simply, artificial intelligence mimics the human brain. He performs tasks according to a given program, analyzes the results and learns.

A teapot with flowers generated from a photo by a neural network

Algorithms are at the heart of the neural network. They process the received requests and try to solve the problem. If you want to teach a neural network to recognize hedgehogs or kittens, load a large number of hedgehogs and kittens into it and indicate to it that these are hedgehogs and kittens. Artificial intelligence is like a child - it will remember what a hedgehog looks like, and in the future, upon request, it will generate pictures of hedgehogs.

At the moment, neural networks are able to create pictures on demand in English, from sketches drawn in advance, as well as mixing several source illustrations.

What neural networks for drawing exist

There are a huge number of neural networks for artists on the market. Create images by text request, source photo or mix existing ones - please. Here are just a few neural networks capable of generating decent images:

• Midjourney;
• GauGAN;
• Disco Diffusion;
• Art breeder;
• DALL-E и DALL-E 2;
• Dreamlike;
• Stable Diffusion;
• NightCafe.

You can learn more about each of these neural networks here. A neural network can create a really cool picture, but for this you have to try. Almost all AIs that do not need to be installed on a computer are paid and the cost of a subscription varies from $10 to infinity.

Red maple drawn by NightCafe neural network

In order for the neural network to generate high-quality art, it is necessary to correctly set text prompts for it and select initial images. Even in the case when you entered the query correctly, you can expect some kind of incomprehensible abstraction or a completely different result that should have been.

What neural networks can and cannot draw

Neural networks are best at drawing landscapes. Whether it's a "colony on Mars" or "mountain ranges at dawn" - you will definitely succeed. Some neural networks are good at drawing portraits and faces of people, but you should be careful here. Since AI best perceives the source in the form of photographs or other art.

The neural network can imitate famous artists and drawing styles. For example, you can try to generate a landscape in the style of Van Gogh, Tim Burton or other authors. Or you can add cyberpunk, steampunk, oil paints, watercolors to this and you get a unique image.

Drawing of a ship created by a neural network

The neural network understands simple queries. Most likely, she will not be able to portray to you “a hedgehog riding a white horse that rides off into the sunset, holding a blade of grass in its teeth.” Or “Pikachu riding a centaur attacking an alien settlement.” But to make an anime-style avatar out of your photo, the neural network will be able to. Queries such as “The big white rabbit eats an orange carrot” are also quite workable.

Kittens with extra limbs. Thanks neural network!

The neural network with hands and anatomy is very bad. In the AI-generated image, cats can sometimes have six legs, while humans can have 15 fingers and several arms growing from a random place on the body. Artifacts also appear in the form of someone's paintings on the finished image, or blurry straight lines. With straight lines, the neural network is also not very good. She can make an awesome landscape with beautiful trees and spoil it by smearing straight branches.

conclusions

The neural network has already been partially able to replace artists. With its help, covers, logos, clips and even games are generated. But at the same time, a person will always be one step ahead. It is the artist who can come up with a plot and draw an illustration the way he sees it in his head. AI will have to learn this for many years to come.

You can generate images using neural networks not only to create something simple. In addition to avatars and a collection of illustrations, you can create a real game with all the art generated by AI, or create a video for your song, magazine cover, advertisement and much more. In addition, neural network processing allows you to create a unique image that belongs to you. So such a picture can be easily completed and corrected in Photoshop or other drawing programs.

Portrait of a girl created by a neural network from a photo

We decided to create several images of Slavic gods using a neural network and this is what happened . If you have created images in a neural network, be sure to try it. Start with something simple, for example, ask the AI ​​\u200b\u200b“How the neural network sees me” and you will definitely succeed.