How Midjourney improves the generation of graphics

Midjourney, along with DALL-E and Stable Diffusion, is one of the most popular image generators using artificial intelligence and machine learning.

Based on a textual description (prompt), it is able to generate hyper-realistic graphics, photos, 3D graphics or illustrations in a variety of styles. How can it help the agency? This is briefly discussed in the article.

How does Midjourney work?

Midjourney’s algorithms generate images from hundreds of millions of graphics published on the Internet. Every few months, the company releases a new model with slightly different features. Currently, version 5 is the standard, and the effects it produces are much more detailed and realistic, and less artistic, than version 4. Five’ does a better job of interpreting natural language and has a higher resolution.

The latest version is currently 5.1, and features even greater consistency, better sharpness, greater accuracy in responding to text commands, and fewer unwanted edges or text artefacts.

However, if the results of our searches are not satisfactory, it is worth comparing them with the results obtained with earlier versions of the program. Simply add the — version parameter or its abbreviation — v to the command prompt, or configure the settings using the /settings command.

How do I use Midjourney?

There is currently no free version available. Midjourney offers three subscription levels, Basic, Standard and Pro, starting at $8 per month.

The service is accessible via a bot on the official Midjourney server on the Discord platform. We can use it by sending a direct request to the bot, or by inviting the bot to our own server on Discord. To generate an image, use the /imagine command at the command prompt. The algorithms will generate four images. We can use the U (upscale) button to enlarge the selected one. We can also generate new iterations of the selected variant by pressing the V (Variations) button. If you want the algorithm to generate images from scratch for a given prompt, click on the arrow button (refresh icon).

Speed is a major advantage of Midjourney. Pictures are literally created before your eyes. In fast mode, you can create an image in about 1 minute. In slow (relaxed) mode it takes much longer. However, a good prompt is essential to ensure that this time is not wasted.

How do you create a prompt?

On midjourney.com we find this definition of a prompt:

“A prompt is a short phrase of text that Midjourney Bot interprets to create an image. Midjourney Bot breaks down the words and phrases in the prompt into smaller chunks, called tokens, which can be compared to training data and then used to generate an image…”. (Source: https://docs.midjourney.com/docs/prompts).

To get results as close to our expectations as possible, we need to learn how to create advanced prompts. Such a prompt should specify elements such as:

Subject: character, animal, place, object, etc.

Technique: photography, painting, illustration, sculpture, drawing, tapestry, etc.

Environment: at home, in the garden, under water, in the city, etc.

Lighting: soft, neon, ambient, studio, etc.

Color: light, muted, bright, black and white, pastel, etc.

Mood: happy, calm, melancholic, etc.

Composition: portrait, close-up, overhead, panorama, etc.

Also important is the use of parameters that instruct the algorithms to change the way an image is searched. These are described in detail and illustrated in the ‘User Guide’ section of the Midjourne website. The most important ones are

• Version (– version,–v) – the parameter already described above, which allows you to select the version of the algorithm.
• Aspect Ratio (–aspect or –ar ) – changes the aspect ratio of the resulting image. The aspect ratio is the ratio of the width to the height of the image. Example parameter: –ar 16:9.
• Chaos (–chaos – the higher the parameter, the more unusual and unexpected the generation.
• No (–no) – negative hints, –no plants will try to remove plants from the image.
• Seed (–seed) – the Midjourney bot uses a seed number to create a visual noise field, such as TV interference, as a starting point for generating the initial image grids. Seed numbers are randomly generated for each frame, but can be specified with the –seed or –sameseed parameter. Using the same seed number and cue will produce similar final images.
• Style (–style) – affects how strongly the default Midjourney aesthetic style is applied to tasks. Toggles between versions of the Midjourney model version 4 –style Toggles between versions of the Niji model version 5 Stylize –stylize or –s .
• Tile –tile – generates images that can be used as repeating tiles to create seamless patterns.

Midjourney also allows you to create advanced image prompts. An image URL can be added to the prompt to guide the algorithms to the desired effect. Image URLs are always placed at the beginning of the prompt.

You can also use the /blend command to blend selected images together, and experiment with further image variations using the /remix command.

Midjourney – pros and cons

Midjourney’s advantage is certainly its ease of use. The tool makes it easy to create graphics, even for beginners in the world of AI. The use of parameters allows you to adapt the operation of the algorithm to the expected effects and gives you the possibility to control the effects to a large extent.

The drawback of the tool is still the limited realism. And not just in terms of common errors such as the number of fingers. It is simply that the algorithm can interpret textual queries in unpredictable ways. This means that changing one word in the prompt can result in a completely different image being generated.

Midjourney and agency work

Advertising agencies have been quick to embrace the possibilities of Midjourney.

It is particularly useful for visualising ideas – whether for internal brainstorming or for presenting ideas to clients. In just a few moments we can create an image that allows the client to feel the concept rather than just imagine it. Midjourney allows us to go beyond the limitations of image banks and create the most unlikely graphics or images on demand. It’s also a great tool for creating mood boards. In our experience, it works quite well as an aid to storyboarding – although as consistency can still be a challenge, this tool should be seen more as an aid to the work and a way of getting it done faster.

Generated graphics:

Hyper-realistic image of a simple monochromatic cup of coffee with a latte art pattern at the top. The latte art pattern includes musical notes and a heart. A green cup on a monochromatic pink background. –ar 4:5 –s 750 –v 4 -.

Illustration showing the entire globe in space. A little boy and a little white bear are standing on top of the globe. The boy is holding a candle in his right hand and stroking the bear with his left. 

Transparent spray plastic bottle for detergent filled with many layers of soil, sand, grass and a single small daisy flower. White bacground. Hyperrealistic graphics. –ar 4:5

A hiperrealistic lollipop made of fried chicken, red background –ar 4:5 –s 750 –v 5 –c 50

A bottle of olive oil at yellow background, surrounded by yellow fruit and vegetables, flat com position, monochromatic, essential, hyperrealistic –ar 4:5 –s 750 –v 5 –c 50

A storyboard full color sketch depicting people packing crates of olives onto the ca r’s trunk in an Olive Garden, wide frame, full-body image –ar 7:4 –s 750 –v 5