Midjourney is one of the top neural networks in the world. For about two years now, the platform has been captivating millions of users with its stunning capabilities for creating various visual materials. The results of Midjourney generation can rightfully be called works of digital art.
However, in contrast to its advantages, the neural network is not so easy to learn and can greatly discourage unprepared users with multi-step registration and a complex interface via the Discord server. We have prepared an article for you to help you better understand the use of Midjourney: find out where to start and how to generate images using popular artificial intelligence.
How the Midjourney neural network works
It is important to understand that Midjourney is not just a resource for generating images. The platform uses its own diffusion model*, trained to understand even the most complex queries and create high-quality art based on them.
*Diffusion models are programs that have gone through multi-stage training to recognize and further generate various queries and images. Such models, for example, include the equally popular Dalle-2 and Stable Diffusion. A simplified description of the problem of diffusion models will sound like this: the uncontrolled transformation of information noise into a distinguishable, representative result.
Those who want to get acquainted with Midjourney have probably encountered many “obstacles” :
- Midjourney does not have its own platform, so the neural network works through Discord;
- The resource has long disabled free functions and requires a subscription, prices for which start from $8-10;
- Midjourney has closed access for sanctioned countries, which forces some users to additionally connect a VPN;
- After activation, the user must join one of the channels: in most cases, generation is carried out via a “live queue” in the general chat;
- To use a neural network, you will have to adapt to a complex interface and an editor that works through special commands;
- Interaction with ready-made results also requires some preparation;
Despite the fact that Midjourney has high barrier to entry, millions of users prefer this particular neural network every day. Below is a step-by-step look at all the steps you need to go through to use Midjourney's artificial intelligence.
Step 1: Login to Discord
On the Midjourney main page, you can log into an existing account, or join a Discord server (i.e. register from scratch). In any case, you will be transferred to the information channel in Discord.
Step 2: buy a subscription to Midjourney
As mentioned earlier, using Midjourney requires a subscription: log into your account page (first screenshot from above), and then select the plan that suits you in your personal account.
Step 3: Join the Midjourney Discord Server
Open Discord and follow these instructions:
- In the left sidebar, select the “+” icon;
- In the creation window servers, select “Join server”;
- In the line with the invitation link, insert http://discord.gg/midjourney;
- Click “Join to server.”
Step 4: select a channel tagged #General or #Newbie
On the left side of the main screen of Discord you have access to a list of chats: select one of those that signed #general or #newbie. These channels are designed for those new to Midjourney, so the bot will not generate images on other channels.
You can also create personal servers to which you can link the Midjourney bot. But first, it’s better to use open channels where you can see the results of generating and configuring prompts from other users.
Step 5: use the /imagine command
Interact with the Midjourney bot in Discord using commands (words preceded by a /). Creating an image begins with the /imagine command, after which you can add your text request (prompt). The sequence of your actions will look like this:
- In the message line, enter “/imagine Prompt:” (other options: 1. click on “+” -> “Use applications”; 2. enter the symbol “/” and find the appropriate command);
- Enter your text request in the line marked “prompt”;
- Send a message so that the bot begins to analyze the request and generate a picture.
Step 6: accept the terms of service
Before you continue using Midjourney, you must agree to the terms of service and follow the community guidelines. After you confirm your consent, the neural network will resume creating illustrations.
Step 7: create an image
Typically, creating images does not take more than 1 minute. After the generation is complete, 4 ready-made variations of your prompt will appear, and under them two rows of buttons:
- The letter U allows you to select one of four options for the finished image (the bot will send it as a separate message), and the number next to it corresponds picture number (the order is shown in the screenshot above);
- The button with circular arrows repeats the generation if none of the results suits you;
- The letter V allows you to start creating a variation of one of the received pieces of art. This is useful if you want to see variations of one of the received images (the bot will use it as a source, maintaining the overall style and composition).
< p>When you decide on an option and select one of the images (U1-4), a new set of buttons will appear with functions for enhancing and changing the image:
- The “Vary StrongSubtle” buttons are intended for fine-tuning the art: with their help you can get improved variations of the selected result, but without global edits. The “Strong” request will maximize the improvements, while “Subtle” will make minor changes;
- Buttons labeled Zoom allow you to reduce or expand the canvas boundaries without changing the contents of the original image.
- Using the arrows (←, →, ↑, ↓) you can select the direction in which you want to expand the image. In the new generation, the picture will be supplemented taking into account the content, but without changing the source;
- The ❤ button adds the picture to favorites, and it can be easily found on the Midjourney website;
- The Web icon allows you to quickly open art in the gallery through your personal account.
In new versions of Midjourney, several additional functions have also been added:
- “Vary by Region” is intended to change some a separate part of the image;
- “Upscale” – increases the resolution of the art by 2 or 4 times (depending on the selected option).
Finally, click on the final version and save the work through the RMB menu (on your phone, just hold down the image and select the download icon in the upper right corner of the screen).
Tips for creating quality prompts
If you've ever tried to generate an image using text queries, you've probably encountered very controversial results: the picture didn't look, to put it mildly, the way you imagined.
Do not rush to blame the neural network for a poor-quality model – instead, check whether you composed the prompt correctly. Make sure that when writing a query in Midjourney, you follow these tips:
- Text query length
Much depends on the length of the prompt, but this does not mean that you need to write an entire essay. Remember that longer queries will help the neural network analyze more data loaded into it, which it will not find without your clear and detailed instructions.
- Grammar
Neural network Midjourney does not understand the grammar or structure of prompts, and is not case sensitive – make sure your input makes sense and is not separated by commas, hyphens, or other delimiters.
- Details
The more detailed your request, the better the system will interpret it. In your text, pay attention to details related to the object being created, the background, colors, lighting, perspective, weather, time of day, artistic style and mood of the art – all this will significantly improve the accuracy of the results.
All these tips will help you get used to working with Midjourney and master the entire sector of capabilities of one of the best neural networks in the world.
