Unlocking the Secrets of Midjourney Prompts: A Beginners Guide to Exploring AI Art
Have you ever been struck by the wild creativity of an AI-generated image and wondered how is it created?
Understanding Midjourney Prompts Better
The key to unlocking the full potential of generative art clearly lies in understanding how to give the AI better instructions. Text-to-image software relies on the user to craft short written inputs. These inputs are called prompts, they are instructions given to an AI algorithm to guide the creation process. How well you write your prompt can greatly impact the final outcome of your AI-generated piece.
What this prompt guide will cover
- The role of prompts in generating AI art.
- We provide tips and tricks for better text-to-image results.
- We dissect a Midjourney generated image to learn how AI sees our world.
This mouse image was generated from the prompt:
"Wild mouse dressed like an action hero hunts a cat in yellow field."
As you can see the AI did a pretty good job of making a heroic mouse, but we don't see any cats around, so the algorithm has mysteriously dropped the cat from its reading of this prompt. Yet it does make a 'yellow' field.
One explanation people offer is that text closer to the start of your prompt has a higher weight for any keywords, while this is true it fails to explain why the AI got the yellow wheat field correct; maybe Midjourney just got lucky.
Another consideration is that you get different results by using 'hunting' instead of 'hunts' hunting is an action and hunts is improper English in this use. Be careful how you word things, minor changes can have a dramatic effect.
Obligatory Hand Jokes Here!
An AI art guide wouldn't be complete without some badly drawn hands and this mouse, while not too bad at first glance, has misshapen fingers as you can see circled in this illustration.
PRO TIP: Use the word 'hand' (not hands) when writing your prompt if you are getting misshapen hands or too many fingers. The AI knows what a hand is if you tell it; otherwise it may not give you a nicely drawn paw.
I like the way the text-to-image AI composed this image of wheat with a subtle green undertone, but then it gets weird here in the front with 'wheatgrass' a blade of grass with wheat at the very top. Lastly even weirder still the seemingly random green leaves directly in front of the mouse. I say random, because the leaves come out of nowhere and are not repeated in the rest of the field.
Light and Regular Upscale Options
one of the keys to Midjourney secrets is knowing this issue can happen when you upscale your image. the AI attempts to increase the image resolution by adding new information when you upscale.
Midjourney offers a few settings to try to soften this problem with light and regular upscale options. You should experiment with these settings to help get a feel for how the image quality changes.
Following the Contour of a Dividing Element
If you notice in this image you can see several examples of the AI following the contour of a clearly visible dividing element. For example if you look closely at the mouse's red cape it passes behind a blade of wheat, then 'poof' the cape seems to be cut off too soon when it is visible again on the other side of the wheat stem. Once again you see this odd behaviour from the AI in the mouse's ear. I have circled it in this image.
You can often remedy these kinds of problems with adjustments to the 'style' setting. Lower style settings do less 'in-painting' which can be one culprit. You can choose from low style, medium style, high style, and very high style (like Snoop Dogg 😆 ).
We hope this article has been helpful, please leave us feedback in the comments, we enjoy hearing from you.