Midjourney: How to Use a Neural Network to Create Paintings
November 10, 2022
17 min read
123 638
designguides
Midjourney: How to Use a Neural Network to Create Paintings
author__photo
Yulia Usacheva
Calltouch Blog Editor
Midjourney is a neural network that recognizes text and creates unique images based on it. You can try the technology even if you are far from graphic design and programming - it's simple and free. We tell you how to use Midjourney, share life hacks and useful recommendations.
What is Midjourney
Midjourney is a neural network that generates images based on text queries. It was created by scientist David Holtz afghanistan phone number resource who worked at the NASA research center and the Max Planck Institute.
Midjourney can be considered a technological breakthrough in the field of artificial intelligence. Its closest competitors are Dall-E 2 (a closed neural network) and Dall-E Mini (more of an entertainment one). Unlike them, Midjourney is open and has practical value - it quickly and efficiently converts text into images.
Midjourney is currently in beta testing and is available to users worldwide. This means that anyone can create images from text queries.
How to get access
The beta version of Midjourney is available only on Discord. It is a bot for exchanging voice, video and text messages with different groups (friends, gaming communities or developers). Discord is a popular analogue of Skype, Zoom and similar programs.
To register in Discord, download the software from the official website , verify your email and phone number. Discord also has a mobile version, but to work with Midjourney, it is better to install the program on your computer.
After registering with Discord, go to the Midjourney website and click on the Join the beta button, then select Sign In. Enter your login and password, and you will be automatically transferred to Discord.
The neural network offers two options for operation:
Through the Midjourney server. After registering in Discord, you will receive an invitation link to the neural network. Inside the server are Newcomer Rooms channels, and inside the channels are Newbies chats. To get started, you can choose any chat. If there are fewer people in it than in others, requests will be processed faster, and your work will not get lost.
In chats, you will see the results that the system is generating right now for other users. They will also see everything you do. This is convenient - you can see how people enter text queries, what results the neural network gives them.
Using the "Add reaction" button, you can send your favorite images to yourself in private messages - to do this, select the envelope symbol in the reactions.
Via your own server. To generate images via a separate server, click on the bot in the Discord user list, then click the "Add to server" button. Next, select the server to which you want to connect the Midjourney bot. This method is suitable if you are uncomfortable working in a common space, or if you do not want to see other people's requests.
When you log in through your own server, your work is available to all users, but not in real time and in the general chat, but on the gallery site. Otherwise, the methods of working through the general and separate server are the same - from basic commands to the time it takes to create an image.
How Midjourney Works
To generate an image:
Open any Newbies chat and start typing the command /. You will see a list of available commands.
Select /imagine. This will bring up Prompt, a required part of the command that cannot be removed.
After Prompt, enter your query, which Midjourney will use to generate an image. You can write anything you want, except for forbidden words. It is better to formulate queries in English.
Send a message to the bot and wait - the neural network will create a drawing before your eyes.
In 3-5 minutes you will receive not one, but four images at once according to your request.
Additional commands
/imagine is not the only command that can be used. Users often enter /info to find out how many requests are left, or /settings to adjust the parameters of the work (stylization saturation, image quality, image aspect ratio, or resolution increase).
Other commands include /show to view the ID, /ask to get answers to questions, /fast to switch to fast mode. To use all the commands, you need to buy one of the plans. However, for simple image generation, the free version is enough.
Menu after image generation
Once you have the image, try to improve it. You can do this using the menu under the image. The following tools are available:
U — increase the original resolution. U1 — top left image, U2 — top right, U3 — bottom left, U4 — bottom right. Select the image number to have the neural network increase its resolution.
Rotating arrows. If you don't like any of the options in the proposed series of renders, click this button, and the neural network will generate new images for this request.
V — variations. By analogy with U, each digit means a certain image: V1 — top left, V2 — top right, V3 — bottom left, V4 — bottom right. Select a picture so that the neural network can create several more drawings based on it.
Additional menu after increasing resolution
After upscaling the selected image, you will see an additional menu with new options:
Make Variations - create a new series of images based on your image with increased resolution;
Upscale to Max — maximum improvement of image quality;
Light Upscale Redo - less detailed upscaling;
Web - link to the finished image on the Midjourney gallery site;
Emoji is an assessment of your work or someone else's image, available in the chat on the common neural network server.
Free E-book
23 Actionable Ways to Make Your Marketing Cooler, Faster, More Effective Than It Is Now
Get it for free
Rules for formulating a query
In the free version of Midjourney, you have only 25 attempts. Any action, including increasing the image resolution, is considered a request. To avoid wasting your click limit, formulate your request correctly.
The main thing is to adapt the request to the neural network. Write more specifically and add a description. For example, instead of "Penguin with a car", write: "Penguin is sitting in a red car". Clarifications like "sitting" and "red" will help the system quickly generate what you need.
This is what the image generated by the query "Celtic knotwork, illuminated manuscript, medieval weave" looks like.
Do not use words such as "not", "besides", "except for" and similar. The neural network may not recognize your request and give you an inappropriate image. You will spend free attempts to correct it.
Another important point is language. Midjourney accepts requests in different languages, including Russian. However, if you use the free version, enter messages in English - the neural network recognizes them better.
Midjourney is familiar with specific people (such as famous photographers), time periods, and different artistic styles. You can experiment with styles from Gothic and Expressionism to Realism and Modernism.
List of neural network parameters
Midjourney has many useful parameters that can be added to a text query:
–hd Generates the entire image series in high resolution, so there is no need to increase it via the U command. Use this command if you want a large image, but be prepared for less detail.
–ar. Parameter used to set the resolution of the image sides. For example, if you put –ar 3:4 at the end of the request, the neural network will create renders in a 3:4 ratio. By default, the aspect ratio is 1:1.
–no. Another useful command in Midjourney that removes unnecessary details. If you add –no cabinet (glass, stone, any other detail) to the request, the neural network will remove the element from the finished image.
-s. This parameter is used to stylize the image. It is subjective, depending on how much you like the Midjourney drawing style. The default is -s2000, adjustable from 0 to 60,000.
–video. A command that allows you to record the process of creating an image in video format . The video will be sent to your Discord private messages. To do this, leave an emoji or
on the created image.This is what the image generated by the following user query looks like: "Ultra-photorealistic, filigree, 4k, bedroom in a luxury apartment, night view of a high-rise city, ethereal lighting."
