Top 5 visual neural networks in which you can make pictures

Author of the article: Tatiana Zhakova

Journalist, linguist, teacher of storytelling with 10 years of experience

In 2015, she created and promoted her project about Nizhny Novgorod nnstories.ru, after which she created the author's course "Storytelling: How to Tell Your Story" on its basis. More than 4,000 students have taken the course.

A linguist by training, she is very quick to master new fields. Now she is actively studying screenwriting and storytelling in films/series, about which she writes in the blog of her project.

Literally not a day goes by that we don't see news that there is a new neural network that does something unique. In fact, it turns out that the functions of this neural network are not much better than a calculator, and the quality leaves much to be desired and is definitely impossible to use in work.
Our team is constantly monitoring the situation and trying out all these new neural networks. Subscribers of our channel "Neural Network Makes Content" also join us - we test neural networks together and share the results with each other. Subscribe, by the way.

So, recently we have identified 5 main visual neural networks that have started to make pictures more or less well.

And now let's test the same query and compare the results.

But first, here's a list of those neural networks:

1

Masterpiece
- is a neural network from Yandex, you can download it from AppStore and Google Play
2

Kandinsky 2.0
- neural network from Sberbank. You can generate pictures in Telegram - @kandinsky21_bot
3

Stable Diffusion
- there's a website, there's a Telegram bot @stable_diffusion_bot
4

DALL-E-2
- you can try the service on the website too, or on Telegram @stable_diffusion_bot
5

And, of course, Midjourney!
Works through a bot in Discord, the only paid neural network of these five. The free period was shut down due to fakes.

I advise you to install these five neural nets, test them and learn how to create pictures in them. They work in the same way - you enter a test prompt, and in about a minute you get a visual response.
Let's compare one query that I asked all five programs to make.

The query:
surrealism, 1800s Victorian Steampunk London glamour, gorgeous detailed eyes, by Erin Hanson, JC Leyendecker, Donato Giancola, Charles Vess, Frank Frazetta, rich vivid colors

It's a bit long for the test, but it's what I personally wanted to test.
I attached the results in the same order, but signed for clarity.

What conclusion can we draw?

Of course, Midjourney is doing the best so far. And it is comfortable to use it in constant work, while other neural networks will hardly be able to make pictures for websites, social networks, and other needs. But it has significant disadvantages: for example, the fact that it does not have a free version, and paid may not be so expensive, but only from a foreign card.

Much more interesting are the conclusions about other neural networks. Yes, their results are not ideal and need to be improved. But remember what Midjourney was doing six months ago! Or google it.

In fact, its picture is the level that we will see from other neural networks in the future. Neural nets learn very quickly. And I think that very soon we will have many high-quality, free and even Russian tools for creating pictures.