Midjourney is a neural network that creates colorful images based on your textual descriptions.
Users create images by sending commands to the bot in the Discord messenger, enter the message “/imagine prompt” enter a verbal description of the desired image, after which the user is prompted to choose the best of the four images generated by the program and receive an image in high graphic resolution.
Neural networks entered our lives very quickly and save a huge amount of time on creating ideas, paintings, and in the near future they can replace some creative professions.
The most important Midjourney parameters
Here we will look at the most important query settings and Midjourney itself.
Image resizing in Midjourney. You can change the aspect ratio of your image with the Aspect Ratio command. To do this, you need to specify a command like –ar x:y after the comma. Let’s say you want to create an image with an aspect ratio of 3:4. Just specify:
Or you want to change the aspect ratio to 16:9. Use the parameter:
You can also specify specific height and width values using the –h and –w commands. If you enter just –h or just –w, then the side you didn’t specify will keep the current dimensions.
Note that changing the aspect ratio of a frame works best when the size is a multiple of 64. For example:
–h 512 or –w 384
Midjourney developers recommend not exceeding 512 pixels, otherwise rendering may not be performed correctly.
Stop rendering early or Stop Render Early. Sometimes you need to generate an image that is less developed than the one that is created at the end of the full cycle of the neural network. And we can instruct Midjourney to stop the generation process ahead of time.
A full cycle is 100 renders, so if you want to stop a render when it’s 50% complete, you can use the stop command:
This is a great way to create a soft composition or scene without focusing on the subject.
Increase brightness when scaling or Uplight Upscale. This option automatically brightens the composition when zooming in.
Of course, you can slightly increase the brightness already in graphic editors after generating the image, but in some cases this will not work. Therefore, to get a brighter output image, you can use the –uplight option at the end of the tooltip.
–no operator. In previous versions of Midjourney, negation in prompts did not work correctly. But now this parameter works better and better. It allows you to exclude from the picture those patterns that you do not want to see in it. Just type –no followed by a description of what you don’t want to see in the finished image.
This command is especially useful if you are using a hint that has multiple values. The –no option can be used to refine the hint.
How to make a request for Midjourney
To create a request to a neural network, use the following command:
/imagine prompt [text prompt].
The queries for the Midjourney neural network can be the same ones you use for Stable Diffusion. Prompts can be both simple and complex, including dozens of words at the same time. If you are a beginner, it may be difficult for you to immediately pick up hints to create the image you want. At the same time, there are no clear rules when creating requests for Midjourney. But there are a few tips that will help you compose the prompt correctly and get the expected result as a result of the algorithm:
- the quality and realism of the image generated by the neural network depends only on the text prompt.
- the more specific the hint, the more information you give the Midjourney algorithm to create the image you intended.
- you don’t have to be super-specific: you can just nudge the neural network and see what it comes up with.
For example, indicating the style of the artist or a particular direction in art helps a lot.
Setting the reference weight in Midjourney
Although Midjourney cannot use source images to draw scenes, you can show the neural network a reference – your own image, which the neural network will focus on when creating a new frame. To do this, use the -iw command.
Experimenting with references is very interesting and, sometimes, this is the only way to ensure that the neural network draws the object you need. For example, a person from your environment, about which the algorithm has no idea.
Step-by-step instructions on how to add a reference to the tooltip with weight settings:
- Find the desired image on the Internet and copy its URL.
- Add the URL to the beginning of the hint.
- After the image address in the prompt text, add the image weight parameter. 1 – this is the normal weight, and it is necessary to build on it. Values below one reduce the influence of a particular image on the generation result, while values above increase it.
Use custom neural network queries, not just [brown cat]. If you already have some kind of image in your head, then try to describe it in the prompt in as much detail as possible – so the discrepancy between the expectation and the actual generation will be less.
Just think about how you would describe the desired image to your friend. Imagine that you are telling him. what picture did you see in the store.
Here is a short checklist that will help you form even a complex text query in a couple of minutes:
- Set the size of both the image itself and individual objects.
- Set a theme (for example, the edge of a forest or outer space).
- Use a description of the style (e.g. mosaic, realism, cartoon).
- Try mentioning an artist, photographer, or designer (like Andy Warhol).
Now you know exactly how to correctly make requests for a neural network and you will be able to generate not a set of random objects, but a work that looks like art, or at least the image that you wanted.
Midjourney Request Examples
ultra realistic close up photography of a bouquet, english gardens flowers, realism, flower photography, 4k, ultra-detailed photography, Natural Lighting , style of Philipa Day flowers
ultra-realistic photography of a lush loose bouquet of ranunculus and eucalyptus twigs in a vase, style of tulipina design, style of janne ford floral, style of La Musa de las Flores, wooden table, mix of muted and saturated colors, white background + bokeh + realism + flower photography + 4k + ultra-detailed photography, Natural Lighting, Beautiful Lighting, insanely detailed and intricate, elegant, hyper realistic, super detailed
ultra-detailed ultra-realistic botanical photo close up Pink ranunculus flowers, Natural Lighting, Beautiful Lighting, hyper realistic, SSAA anti-aliasing max-graphics ultra-quality post processing, soft lighting rich colors, bokeh, blur background