Midjourney

What is Midjourney?

Midjourney is an AI image generator that operates through a Discord bot interface as well as a web app (an API version of Midjourney is planned). The process for generating images with Midjourney follows the same fundamental principles as other AI image generators, including the use of prompts to guide the generation process.

One of the unique features of Midjourney compared to other AI image generators is its ability to create visually striking and artistically composed images. This is attributed to the model's specialized training, which enables it to produce high-quality images with specific artistic parameters (more about this in "Advanced Prompts" > "Parameters").

You can experiment with the Midjourney Bot in the Learn Prompting Discord or in the official Midjourney Discord server.

Basic Usage of Midjourney

The basic prompt anatomy with Midjourney is /imagine prompt: [IMAGE PROMPT] [--OPTIONAL PARAMETERS].

For example: /imagine prompt: astronaut on a horse

Example with parameters: /imagine prompt: astronaut on a horse --ar 3:2 --c 70 --q 2 --seed 1000

In this basic example, the following parameters have been used:

--ar 3:2 sets the aspect ratio of the image to 3:2

--c 70 adds a chaos value of 70 to allow Midjourney to interpret the prompt more freely(chaos value range: 0 - 100)

--seed 100 sets an arbitrary seed value which can be used to re-render or re-work an image later.

(learn more about Midjourney parameters in "Advanced Prompts" > "Parameters")

Advanced Prompting With Midjourney

Advanced prompts in Midjourney make use of parameters and special prompting techniques supported by the Midjourney algorithm.

Multi Prompts

Midjourney interprets your prompt holistically by default. Using a double colon :: tells Midjourney to interpret each part of a prompt separately.

Example:

/imagine prompt: astronaut and horse

/imagine prompt: astronaut:: and horse

Image Prompts

By uploading an image to Discord and using its URL in a prompt, you can instruct Midjourney to use that image to influence the content, style, and composition of your results. Example: Astronaut (Source: Wikipedia)

/imagine prompt: [image URL], impressionist painting

Parameters (v4)

The following parameters are supported by Midjourney's latest model (v4).

Aspect Ratio:

--ar [ratio] changes the default ratio (1:1) to a new ratio (currently the max. supported ratio is 2:1)

Example: astronaut on a horse --ar 16:9 and astronaut on a horse --ar 1:2

Chaos:

--c [value] sets a chaos value that determines how much Midjourney varies the prompt; the higher the chaos value the more unusual and unexpected the results and compositions (range: 0 - 100)

Example: astronaut on a horse --c20 and astronaut on a horse --c 80

Quality:

--q [value] defines how much time will be spent generating the image, thus increasing the quality. The default setting is "1". Higher values use more of your subscription's GPU minutes (accepts values ".25", ".5" , "1" and "2")

Example: astronaut on a horse --q .5 and astronaut on a horse --q 2

Seed:

--seed [value] sets a seed number that defines the starting point (noise field) for the image generation. The seeds for each image are generated randomly when not specified with the seed parameter. Using the same seed number and prompt will produce similar images.

Example: astronaut on a horse --seed 123

Stylize:

--stylize [value] or --s [value] influences how strongly Midjourney applies its artistic algorithm. Low values produce images that closely match the prompt, high values create very artistic images that are less connected to the prompt. The default is 100, value range is 0 - 1000. (Note: you can use the /settingscommand to change the default stylize value from "🖌️ Style Med" (=--s 100) to "🖌️ Style Low" (=--s 50), "🖌️ Style High"(=--s 250) or "🖌️ Style Very High" (=--s 750))

Example: astronaut on a horse --s 50 and astronaut on a horse --s 900

Version:

--v [version number]or --version [version number] let you access earlier Midjourney models (1-3)

Example: --v 1, --v 2, and --v 3

Parameters (previous models)

Same Seed

--sameseed: while the --seed parameter produces a single noise field applied across all images in the initial grid, the sameseed parameter applies the same starting noise to all images in the initial grid so it will produce very similar images.

Example: astronaut on a horse --sameseed --v 3

Tile

--tile generates images that can be used as repeating tiles to create seamless patterns for fabrics, wallpapers and textures (only works with models 1 - 3)

Example: astronaut on a horse --tile --v 3

Video

--video creates a short movie of the image grid being generated. Reacting with the ✉️ emoji lets the Midjourney Bot send you a DM with the link to the video.

Example: astronaut on a horse --video --v 3

Conclusion

Midjourney offers a large arsenal of tools and parameters to add to your image prompts and generate impressive outputs. Familiarizing yourself with these techniques will help you put your new prompt engineering skills into practice with a popular model.

FAQ

What is Midjourney?

Midjourney is a popular AI image generator that can be accessed through a Discord bot interface or a web app.

What kinds of parameters can I include in my Midjourney prompts?

Midjourney supports many different parameters that can be used to specify your image prompts. For example, you can explicitly determine the aspect ratio of the output and influence how stylized you want it to be.

Links

Official Midjourney Documentation

Sander Schulhoff

Sander Schulhoff is the Founder of Learn Prompting and an ML Researcher at the University of Maryland. He created the first open-source Prompt Engineering guide, reaching 3M+ people and teaching them to use tools like ChatGPT. Sander also led a team behind Prompt Report, the most comprehensive study of prompting ever done, co-authored with researchers from the University of Maryland, OpenAI, Microsoft, Google, Princeton, Stanford, and other leading institutions. This 76-page survey analyzed 1,500+ academic papers and covered 200+ prompting techniques.