The developers of the sensational Midjourney clos free access to the service at the end of March. Now you have to pay from $10 monthly to use the neural network. Therefore, we have collect for you 3 free analogues that generate a wide variety of pictures quite well.
This is not a rating, but just a selection – each of the neural networks has its pros and cons. We will analyze them so that you can choose a service to your liking. And you will not ne a VPN to connect to them.
Why Midjourney Was Shut Down
First, a few words about what happen to Midjourney. There are two versions – the official one and what is being said on the Internet.
Let’s start with the second one. The neural network became a hostage to its own popularity. The publication of the fifth version caus a furor – the neural network learn to draw hands and the quality of images significantly improv, it became more difficult to distinguish them from real photos. The new features were immiately put into action – users began to generate fake photos, which l to misinformation. Thus, images of Pope Francis in a fashionable down jacket and “photos” of Donald Trump’s arrest appear on the network . The deepfakes were so realistic that some took them at face value.
A week later, the opportunity for free access was clos
The official version is extreme demand and abuse. The developers deny the version that access was clos due to fake images. According to Midjourney founder and CEO David Holtz, users creat many disposable accounts to use Midjourney. Such a load l to service failures. The developers said that the measure is temporary and access will be clos until new security protocols are implement. In the meantime, you can use the neural network for $10 per month. Or switch to similar ones.
We have select 3 neural networks, popular and not so popular.
To objectively evaluate their capabilities, we have compil 3 dissimilar queries
an astronaut paints a picture on Mars with brushes on canvas, planet Earth is visible in the background, detail photo – detail request;
the girl rais her palm – we are testing the skill of drawing hands;
love – we test the ability to understand abstract requests.
Let’s see what came out of this.
Kandinsky
Kandinsky 2.1 is a development of a Sber division. The neural network understands requests in 101 languages, including Russian — a unique feature that no analogue has. The service is available:on the ruDALL-E website – several generative models of Sber are present here;
using the voice assistant “Salute” from Sber – it has a special function “Turn on the artist”;
in the telegram bot .
Learn how to purchase a business-specific mobile phone number data using our mobile phone number data. All you need to do is visit the business japan phone number data mobile phone number data website, click on the list icon, enter the required business and industry details and finally make a request for the mobile phone number data in the format you want to buy.
Generating images with Kandinsky is pure creativity and a fascinating process. You can choose any of 24 styles for the picture, from anime to Khokhloma, and the neural network will illustrate. The service can not only draw, but also correct what has been drawn. There are 3 generation options available:
According to the text description – a simple mode for creating pictures
Combining multiple images. This is convenient when you ne to combine several different styles in a picture – the neural network will do everything for you.
Reworking the image. If you don’t like some detail, the eraser is at your service – you can erase any area and ask Kandinsky to rraw it.
The eraser is one of the best solutions! Other neural networks definitely lack it. For example, we ask Kandinsky to draw “a black cat sitting at a table, drinking coffee and eating a croissant on the veranda of a restaurant in Paris.” On the left is the original version, where the cat looks very strange. The eraser fix it.
Images generat by the Kandinsky 2.1 neural network
Now let’s move on to our task for neural networks and see how Kandinsky did. Let’s remember that we had 3 requests.
Kandinsky generates only square images, but if you move the frame to the right/left and send the request again, the neural network will finish drawing the image, and we will get the size we ne.
1. An astronaut paints a picture on Mars with brushes on a canvas, the planet Earth is visible in the background, detail photo. It turn out well, although the landscapes are rather strange and there is something wrong with the perspective.
This is how the Kandinsky 2.1 neural network sees the astronaut-artist
2. The girl rais her palm – we decid to demonstrate how the neural network copes with different styles and generat a request in the spirit of anime. It turn out interesting, but Kandinsky’s hands are not given yet.
Girl with rais palms in anime style by Kandinsky 2.1
3. Love (in the Renaissance style) Kandinsky leads and email marketing: a relationship to cultivate depict like this. The characters seem to have come from the pages of the works of Italian poets Francesco Petrarca and Giovanni Boccaccio. Definitely a like! If you choose the “no style” generation, the result will, of course, be different, more philosophical and abstract.
“Love” in the Renaissance style perform by Kandinsky 2.1
Pros and cons of Kandinsky 2.1
Pros:
ease of use;
completely free;
You can reproduce artist styles or 3D renders;
does not require phone number qa registration;
supports 101 languages, including Russian;
it is possible to rraw or correct parts of the image;
The itor field looks like an endless board – you can generate wide-format images.
Cons:
It happens that the service freezes if the server load is high
the neural network does not understand complex, detail queries well;
pictures for one request are quite similar;
often generates something other than what is ne. You have to spend a lot of time composing the correct query.
What’s the result? Yes, Kandinsky lags behind Midjourney, but stylistically it generates very similar images, and the service is free. If you spend an hour or two with it over a cup of tea, you can get very decent results.
