At first, I was annoyed and disappointed. My expectations were not met by Midjourney.
But then (after several trials) I understood that the way I input the prompt was not understood by Midjourney. Who was to blame? Me? Probably not, however, on my modest experience with machines, I found that they tend to ignore the broad language, meaning, I had to be more specific.
I realized that there were different factors about my prompt/s that caused Midjourney to create weird, scary, unpurposely funny, or undesirable images.
Things started to change when (deep in my AI art rabbit hole) I manage to classify the mistakes I got based on the prompts. Here are some teachable moments.
- This is what happens when you have a Narrow vision
If you have a very well-defined vision in your mind but you aren’t prone to deviation, in other words, your vision doesn’t have much room for interpretation, you might set yourself up for disappointment. Like when I asked Midjourney to create a picture of “skin texture”, with the idea that people could use the picture to apply to 3D models. I got brown gradient lines with texture on top and a face that had a grainy texture.
2. How can you recover from a Poor prompt description?
Sometimes descriptions are not enough to get the type of image you want. AI art generators have keyword descriptors and special codes that help them filter possibilities and add the details you desire to the picture. Due to the frustration of having to read the manual with all the codes, people have created prompt generators that take into consideration the different aspects that go into creating an image. Prompt generators consider the art medium, the technique, the artist or artists that created it (yes you can make two artists collaborate on your creation), the period, the lighting, the type of focus, etc, etc. So one of the first times I created a pattern I was taken by the fact that the pattern was not seamless meaning that it didn’t have continuity. The solution to that was to add the code –tile after the description.
Learn more on how to create effective prompts with these videos.
3. Guess what happens when the AI is untrained on the topic
In some cases, the AI art generator might not have enough data to create the type of picture you want so the result is not clear. I call it an aberration. That was my conclusion when I asked to create a picture for an “Apple computer website”. I got 4 different variants of an apple drawn in different styles (reminiscent of the 80’s style).
4. From multiple interpretations the AI picks the least obvious
In some cases, the AI misinterprets a prompt that you might deem obvious and sometimes creates a literal interpretation of what you prompted. One example is the “Family tree”. Most of us would picture a diagram made of boxes with people’s names and dates inside linked with arrows to signify relationships and their offspring repeated multiple times from top to bottom. Mid journey created a tree with a family around it also created.
Using the right language when interacting with AI is important. You need to be precise, understand the code, be open to surprises, and be willing to try again.
At the end of the day writing a prompt with the AI art generator is an exercise of communication similar to communicating with another person. The difference is that the AI has way more information in its hands than we have, but they are not curators we need to be curators for them.
This moment in time we are witnessing is the greatest technological advance. The work of AI is already changing us forever in ways we can’t even imagine.
How do you think these AI experiments will change us?
Hold your thought, I will talk about them in Part 3.
Part 3 of the article coming soon.