News about neural networks has been appearing frequently lately. Artificial Intelligence (AI) has learned to write lyrics and music, create videos, and draw pictures. Representatives of creative professions are already using neural networks in their work.

We will tell you how neural networks can be applied in different spheres. What is needed for AI to understand the task correctly. We will show how to draw a picture by the example of Midjourney.

Why to create a picture in a neural network

Neural networks can be useful for bloggers, illustrators, designers. And other specialists, who create or use visual content in their work.

According to many experts, neural networks will not completely replace artists. At least in the near future. In the works of artificial intelligence individual creativity, authorial style and atmosphere are not enough. At the same time the arts are of high quality and detailed. That’s why neural networks, including Midjourney, can become reliable assistants for creators.

What you can use neural networks for:

• Creating article covers and posts. AI can create a lot of content in a short amount of time. This is important if you publish frequently. Visual design can be entrusted to a neural network and free up time for other tasks.
• Logo design. Of course company’s mission and values are hard to “pack” into a few lines of a neural network query. So the result is likely to require some fine-tuning. But you may get the basic variants.
• Checking ideas in design. A neural network spends a couple of minutes on work that takes a human to do in a few hours. Artificial intelligence can be used to check the combination of colors, fonts, proportions, styles. Will come in handy for both web designers and interior designers.
• Creating illustrations. With the help of neural networks one can create commercial and artistic illustrations. For example, to draw pictures for advertising materials or sites; to decorate books.
• Searching for inspiration. Even if an image cannot be used immediately without modification, it will give the artist a general idea. The creator will develop it into a full-fledged artwork.

How to make a picture in Midjourney neural network: requests, words, description

This point we’ll start with the theory – talk about the general principles of building queries and special commands for communicating with Midjourney. Then we give you an example of how to build a query. Then we will give step-by-step instructions on how to generate art using a neural network.

The general principles of making queries:

• Use /imagine prompt at the beginning of each query.
• Separate semantic parts of the query with two colon characters “: :” – do not forget to put spaces before and after each character.
• Use special commands – technical parameters of the image can be set with commands. Put a “-” before the command.

What commands can be used in queries:

• ar – sets the aspect ratio of the picture. If the parameter is not specified, the image is rendered in a 1:1 ratio. You can replace it with, for example, 3:2 or 7:4.
• video – neural network saves video and sends it in personal messages in mp4 format if requested.
• v – allows you to choose the version of the neural network you will work with. Now Midjourney has four versions.
• chaos – the degree of diversity and randomness of the results. After the command, specify a number from 0 to 100.
• stop – stop creating an art at a certain stage. Specify a number from 10 to 100 after the command. It corresponds to the percentage of the finished art.
• s – stylization level. Denoted by a number from 700 to 20 000.
• q – level of quality and detail. Specified by numbers from 1 to 4.

To receive the desired result from the neural network the request must be correct and detailed. The more accurate you are in specifying image parameters, the better the neural network will understand the task.

It is recommended to form requests according to the scheme:

• The first part – the subject, person, place or situation you want to represent.
• The second part – details and circumstances. Here you can specify a name of artist or photographer, whose work you want to style your art. The neural network knows styles of painting and architecture, distinguishes historical epochs, times of year and day, etc. That is, you can prescribe any plot for the future picture.
• The third part is the technical parameters. Use the commands to set the image format, the level of stylization and detailing.

An example of a ready request:

/imagine prompt young girl : : flowers in the field : : beautiful : : morning : : girl collects flowers : : -video -v 3 -ar 3:2 -q 2

So we have asked the neural network to draw a girl who is picking flowers. The process of generating the picture will be saved on video. The third version of Midjourney will work. The result is an image of improved quality with an aspect ratio of 3:2.

These are the options proposed by the neural network on this request. You can refine and detail the pictures.

How to make a picture with the Midjourney neural network

1.Open midjourney.com and click “Join the Beta”.

2. Neural Network is available only through the chatbot in the service “Discord”. So you need to log in or register.

3. You will receive an e-mail asking you to confirm your registration.

4. Open any chatbot with the name “newbie”.

5. Click on the grid icon to open the chat branches. Choose one where the last activity was a long time ago. Then you will not have to search for your requests in the general feed, and arty will not be lost.

6. Compose your request. Use online translator if necessary.

7. In the message window select the command /imagine prompt – it will appear in the automatic prompts. Then insert the query text, add commands and send your message.

8. Acknowledge your agreement with service rules.

9. Generation of images can take from several seconds to several minutes. Depending on the quality and detail parameters you set.

10. Neural network will offer you 4 basic options of the picture.

11.To make your favorite option more detailed and get a better result, press the button with the letter “U” and the number of the image.

12. 12. The finalization will take a couple of minutes more. To save the picture, click on it and click “Open in Browser”. Then save from a separate tab. That way you’ll get art without losing image quality.

13. The neural network can create four additional variants of each picture. Click on the button with a “V” and the number you want.

14. You can change the settings of the quality and version of Midjourney directly in the chat. To do this, send a message with the command /settings and select the desired icons in the list that appears.

15. Neural Network can send the created images and videos of their generation to you in private messages. To do this, send an “envelope” emoji in response to a post with the desired art. You can find it by the word “envelope”.

Tips on using neural networks or how to get the right result

If you plan to use Midjourney regularly, it’s better to create your own server in Discord. You can invite a neural network chatbot there and communicate with him personally, not in general groups.

• 1.Log in to Discord and click “Add Server”.
• 2.Open the Midjourney chatbot and click “Accept Invitation”.
• 3. In any chatbot group, find the active link with the name of the neural network and click on it. Add the bot to your server.
• 4. You can send requests.

The result of neural network depends on the quality of request. There are several ways to get images that will be exactly right for your purpose.

How to get the best results from neural networks:

• Avoid vague wording. Midjourney has broad capabilities, but it’s still artificial intelligence. So don’t use complex sentences and artistic descriptions in your queries. Select the array of phrases that best describe the task and spell them out one by one. Example: instead of “a flock of birds flying in the rays of the setting sun,” write “flock of birds, sunset, birds flying.”
• Add details. Write the details of your art in the query. Describe the time of day, weather, historical era, style of picture, prevailing colors. If the request is a few lines long – no big deal. But you have a better chance of getting the results you need.
• Do not use negation (if possible). The words “without”, “except”, “no” are not always perceived by the neural network, and can give the opposite result. Example: Instead of “warm shades other than red” write “in yellow and orange.
• Try different versions. If the neural network does not accept the commands, change the version in the settings. Sometimes there is a problem of version incompatibility with some commands.
• Use services for writing queries. These are resources that can help you formulate the correct request by using commands from ready-made lists. Examples of popular services: Promptomania and Noonshot.

With Noonshot the users specify a description of the future picture, choose styles and effects and then copy the ready request

Conclusion

Neural networks can benefit everyone who works with visual content. AI can be used to create illustrations, design articles and posts with pictures.

Midjourney is one of the most popular neural networks for creating images. It is available through a chatbot in Discord. If you want to get good results, you need to formulate your request accurately, precisely, and correctly. Special query commands and services will help. Midjourney will suggest four picture variants. Then each can be refined and detailed.