Hello there Midjourneyers!
Creating images using AI is a wonderful journey. It is fun, it is exciting and it is also useful. But getting the prompt right is only a part of the game. If you truly want to let your imagination shine, you need to get to know the tool a little bit better.
Last week we introduced ourselves to the concept of PrompCrafting. Today we are going to explore further.
We are going to learn about the Parameters part of our prompts. Specifically the versions of the models available through Midjourney.
Today we are going to learn about:
Midjourney Parameters
Different Midjourney Models
How to use them
What are Parameters?
Parameters are options we add to our prompts that influence how an image is generated. By using parameters you can change the model (version of the algorithm) used to generate the image, aspect ratio, rendering quality, amount of stylization, randomness, and more.
You can change some of the Parameters by using the /settings command.
you type it in and hit enter a visual menu will pop up.
You can access some of the parameters here. But you can also add them manually to the prompts.
Let’s look at how to do it and what these parameters mean!
Adding Parameters
To add parameters to your prompts you just need to type “--” followed by the parameter and your chosen value.
In this example prompt I am using two separate parameters. --ar, which stands for aspect ratio (in this case I have chosen 16:9), and --v, which stands for the model I am using to generate images (in this case it is the version 5 model).
Parameters should always be placed at the end of your prompts.
Today we are going to focus on the --v parameter. Don't worry we will look at other parameters in the future!
Model Version
This is arguably the most important parameter and will influence your generations drastically.
Currently, there are 5 versions of the main Midjoruney algorithm (--v 1,--v 2,--v 3,--v 4,--v 5) you can use, as well as some alternative model versions. (--niji, --hd, --test, --testp)
Let’s have a look at these.
I’m going to use a simple prompt to compare the versions:
I used the same simple prompt - an elderly man with glasses, for all of these images and just changed the version parameter. You can observe how Midjourney has evolved through its versions 1 to 5.
If you are looking for a certain type of image sometimes an older version might capture the aesthetic you are going for better. But overall the progression forward is noticeable with every new model the images get better.
Let’s talk more about the alternative versions.
Niji
This is an alternate model trained specifically to capture the anime aesthetic.
You can enable Niji by activating it in the /settings menu or by adding --niji to the end of your prompts.
Here are some V5 images compared to Niji:
Prompts used:
A cyberpunk anime girl --ar 9:16 --v5 / A cyberpunk anime girl --ar 9:16 --niji
Prompts used:
An elder rat samurai by jordan grimmer, Jean Leon Gerome, j c leyendecker. --ar 2:3 --c 15 --v5 / An elder rat samurai by jordan grimmer, Jean Leon Gerome, j c leyendecker. --ar 2:3 --c 15 --niji
Prompts used:
beautiful woman actress kissing a bot, full body --ar 16:9 --v5 / beautiful woman actress kissing a bot, full body --ar 16:9 --niji
If you are going for the anime look niji definitely is the model to use. But it can also be useful in other scenarios if you aren’t looking for photorealism.
Test and Testp
Both of these are earlier test models which were released before the Version 4 algorithm. The Test model leans more on the creative side while Testp is oriented toward photorealism.
You can enable these by activating MJ Test or MJ Test Photo in the /settings menu or by adding --test or --testp to the end of your prompts.
Let’s compare these to V5:
Prompts used:
a tiny cute photorealistic piglet sitting covered with a red blanket, drinking hot chocolate from a blue mug on a backdrop of snow with rosy cheeks --ar 9:14 --creative --test // --v5 // --testp
V5 takes the win on prompt interpretation, but both the test models have generated creative and cute images with quite a unique look. While version 5 is great if you know exactly what you want, these test models and earlier versions of Midjourney’s main algorithm can surprise you with some great and creative outputs.
Now let’s use a dictionary description of an elephant as a prompt and let’s see how the models interpret it differently.
Prompts used:
a very large wild animal that lives in Africa and Asia. It has thick grey skin and a very long nose called a trunk. Wide angle shot --ar 9:16
--test // --v5 // --testp
As you can see V5 understands we are talking about an elephant, but the test models give a bit wilder results. While Testp is close to the real thing and looks photorealistic it’s still something else. And the image generated using the Test model is something else.
If you want to reimagine existing things in a new way test models might give you better results.
HD
This is an alternative earlier model that creates larger images but is less consistent. Can be used to create abstract images.
This model works with versions up to v3.
You can enable this model by activating v3 or lower in the /settings menu and by adding --hd to the end of your prompts.
Prompts used:
a photorealistic portrait of a woman --v 3 --hd / the invisible hand --v 3 --hd
Prompts used:
the concept of time and everything --v 3 --hd / he concept of time and everything --v 3 --hd
As with all the earlier models. It’s not as easy to get exactly what you want. But it does feel like the model has its own defined style. This won’t get you photorealistic results, but it might be the right fit for you if you like the distinct style of early Midjourney algorithms.
Have Fun!
Take your time exploring the different models available through Midjourney. Depending on what you create you might find that an older model is exactly what you have been looking for.
The different models also respond to the way you prompt them differently. If you are not as experienced with prompting you might find versions 3 and 4 easier to use than version 5.
There are no rules that are set in stone. Don't be afraid to experiment.
Got any questions, suggestions, or anything else to say? Reply to this e-mail or leave a comment on Substack!
See you again soon! :)