How Does Midjourney Work – In-Depth Look


If you click on a link and make a purchase, I may receive a small commission. As an Amazon affiliate partner, I may earn from qualifying purchases.
Read our disclosure.

Key Takeaways

  • Midjourney uses large language and diffusion models to generate high-quality images.
  • Users use text prompts to guide the image generation process.
  • While Midjourney offers free use of its AI art generator, due to extreme demand, its availability is highly unlikely.
  • The generator offers relax and fast image generation modes. Fast is only available with a paid subscription.
  • Midjourney offers Commercial licenses for subscribers, but don’t forget to read about copyright rules surrounding AI art.
  • Midjourney is in constant development.

My Experience

“Midjourney generates some of the best-looking AI art. The user experience is definitely not the best, and it’s all “thanks” to the way Discord works. If you are familiar with Discord, then using Midjourney is much easier. To make Midjourney enjoyable, it’s best to get the Standard plan. The free plan of Midjourney is basically unavailable as there’s too much demand from paying subscribers. While Midjourney is one of the best AI art generators available, you might get more personal results by locally installing Stable Diffusion and AI art models (specific to your taste) from Civitai.”

How Midjourney Works

Midjourney is an AI image generator that uses large language and diffusion models to create unique images based on text prompts. While we don’t precisely know how Midjourney works, how they train their AI art models, or what datasets they use, it is known that Midjourney employs a combination of machine learning technologies.

The large language model helps understand the meaning of the prompts and converts them into numerical vectors. These vectors guide the diffusion process, where random noise is gradually added and reversed to generate high-quality images.

Diffusion models (especially Stable Diffusion) have gained popularity recently, allowing computers to learn from training datasets and recover original images by removing noise. Midjourney leverages this technique to produce visually appealing artwork by denoising random images.

Image showing the reverse diffusion process.

When a user inputs a text prompt, the AI image generator begins with a field of visual noise, similar to television static. The noise is gradually subtracted through latent diffusion, resulting in an image representing the objects and concepts described in the prompt.

It’s important to note that the image generation process takes time (GPU-minutes in Midjourney terms). Waiting a minute or two allows the AI model to go through enough denoising steps, resulting in a fully developed image. Ending the process prematurely yields a noisy image that hasn’t undergone sufficient denoising.

How Midjourney Text Prompts Work?

Midjourney has built its own text prompt parameters that differ from what you tend to use in every other AI art generator. The image generation in Midjourney starts by first joining one of the #newbie channels in their Discord server.

Type /imagine to start the text prompt in Midjourney.

By typing /imagine into the Discords chat bar and then your text prompt, for example,

90s anime still, asuka langley soryu wearing a red, neon genesis evangelion, watercolor illustration --ar 2:3 --niji 5 --style scenic


…you can start the image generation process.

Midjourney supports negative text prompts the same way as Stable Diffusion (especially 2.x versions), but the text prompt begins with --no (and what you don’t want to see), for example, --no flowers.

Check out: The Best Midjourney Prompts

Using images

Midjourney is especially powerful because it can fetch any image online to be part of the image generation process. You can also upload images to Discord and then use image prompts to generate an image.

You can simply type or paste the web address of an online image with extensions like .png, .gif, or .jpg. These image addresses are placed at the front of the prompt. It’s important to note that prompts must have either two images or one image and additional text.

To obtain an image URL, users can right-click or long-press an image and choose “Copy Image Address” in most browsers.

If you want to use your own personal images in a prompt, there is a method to upload the image to Discord. By clicking the Plus sign next to the message input area, you can select “Upload a File” and choose the desired image.

Aspect Ratios

1:1 is the most common aspect ratio you see when it comes to AI art. With the release of V5, you can now use any aspect ratio that comes to your mind. The default aspect ratio is 1:1, but you can change it with the --ar or --aspect parameter. --aspect parameter uses whole numbers such as 472:1000 instead of 4.72:10.

The aspect ratio may change when you upscale the image.

Common Midjourney aspect ratios:
--aspect 1:1 Default aspect ratio.
--aspect 5:4 Common frame and print ratio.
--aspect 3:2 Common in print photography.
--aspect 7:4 Close to HD TV screens and smartphone screens.

Midjourney Prompt Guide

If you know how other AI art generators work, especially Stabel Diffusion, the same text prompts don’t work in Midjourney. To get the most out of Midjourney, you should get familiar with all the parameters Midjourney offers, as there are a lot of them.

With the text parameters, you can better guide the generator to create art specifically to your requirements and taste.

The more specific you are with your text prompt while making sure you understand and know how to use Midjourney’s parameters, you will get the best possible results.

How to Use Midjourney


Here is a step-by-step guide to getting started with Midjourney:

  1. Log in to your Discord account: Access Midjourney Bot on Discord through the web, phone, or desktop apps. Make sure you have a verified Discord account.
  2. Subscribe to a Midjourney plan (optional): Visit Midjourney.com/account and sign in with your verified Discord account. Upgrade or downgrade your plan as needed. For pricing and more information, check Subscription Plans on the website.
  3. Add the Midjourney server on Discord: Press the “+” button at the bottom of the server list on the left-hand sidebar in Discord. Click the “Join a Server” button and paste or type http://discord.gg/midjourney.
  4. Join the appropriate channel: On the Midjourney Official Server, select any general-# or newbie-# channel visible in the left sidebar. On other servers that have invited the Midjourney Bot, follow the instructions provided on your server for using the bot.
  5. Use the /imagine command: Interact with the Midjourney Bot by using commands. The /imagine command generates a unique image from a short text description called a prompt. Type “/imagine prompt:” or select the /imagine command from the slash commands pop-up. In the prompt field, describe the image you want to create in a simple and concise sentence. Send your message to initiate the image generation process.
  6. Accept the Terms of Service: The Midjourney Bot will display a pop-up asking you to accept the Terms of Service. You must agree to the terms before any image is generated.
  7. Wait for the job to process: The Midjourney Bot takes about a minute to generate four image options. It utilizes powerful Graphics Processing Units (GPUs) to interpret and process each prompt. Each time you create an image, you use some of your subscription’s GPU time. You can check your remaining Trial time using the /info command.
  8. Upscale or create variations: Once the initial image grid is generated, two rows of buttons will appear. The U buttons upscale the image, generating a larger version with more details. The V buttons create slight variations of the selected grid image, generating a new image grid with a similar style and composition. The 🔄 (re-roll) button reruns the original prompt, producing a new grid of images.
  9. Create variations or favorite your image: After upscaling an image, a new set of options will appear. The “Make Variations” option creates a variation of the upscaled image, generating a new grid of four options. The “Web” option opens the image in your gallery on Midjourney.com. You can also tag your best images as favorites to easily find them on the Midjourney website.
  10. Save your image: Click on the image to open it to full size. Right-click and choose “Save image” to save it. On mobile devices, long-tap the image and tap the download icon in the top right corner. All images are immediately available to view on midjourney.com/app.

Note that you can’t use Midjourney without Discord. You can use Midjourney on your own Discord server, but you can’t use it without Discord. Installing Midjourney locally to your computer isn’t possible as of right now.

You can use Midjourney privately by installing Midjourney on your own Discord server or by getting the Pro subscription plan that has a Stealth image generation option.

How to Use Midjourney For Free

While Midjourney offers free use of its AI art generator, you will likely see the following message when typing a text prompt: “Due to extreme demand we can’t provide a free trial right now. Please /subscribe or try again tomorrow.”

This means that while you can use Midjourney for free, there’s too much GPU demand (image processing, in this case) from paying subscribers. In other words, you can’t use Midjourney for free.

Fast And Relax Image Generation In Midjourney

Midjourney offers two modes, “fast” and “relax,” for image generation. The “fast” mode prioritizes immediate access to a GPU, making it the highest priority processing tier, albeit at a higher cost. On the other hand, the “relax” mode places users in a queue based on their usage history in relax mode.

The basic plan grants users 200 fast GPU-minutes per month, while the standard plan provides 15 fast GPU-hours per month, and the pro plan offers 30 GPU-hours per month.

To put these numbers into perspective, approximately 60 image generation or upscale commands, and roughly 200 image variation commands, can be executed within a single hour.

It’s important to note that these numbers are subject to change as they are part of ongoing experimentation.

The standard and pro plans include relax mode, which can be activated using the command “/relax” and deactivated using “/fast.” The image generations do not consume fast processing time when in relax mode.

However, it should be noted that the basic plan does not include relax mode, and the “Max Upscale” feature is not available in relax mode.

In addition to the allocated plan resources, users can purchase Rollover Fast Hours. These hours have no expiration date but require an active subscription to be utilized.

You can use the /info command to check how much Fast time is left for the subscription period and other statistics about your subscription.

How many GPU minutes do the generations cost?

On average, one job takes roughly one minute of GPU time to finish an image. Upscaling an image or using nonstandard image ratios (other than the common ones) may take more than one minute to finish an image. Variations and lower-quality images take less than one minute.

On average, you can generate 60 images with one GPU-hour (subject to change). If you rank image pairs at Midjourney you can earn free Fast GPU-hours. The top 2000 daily image raters receive an hour of free Fast GPU time.

Feature image credits.



Digital Artist

I’m a digital artist who is passionate about anime and manga art. My true artist journey pretty much started with CTRL+Z. When I experienced that and the limitless color choices and the number of tools I could use with art software, I was sold. Drawing digital anime art is the thing that makes me happy among eating cheeseburgers in between veggie meals.

More Posts

Thank You!

Thank you for visiting the page! If you want to build your next creative product business, I suggest you check out Kittl!