It is difficult to predict exactly what Midjourney will make of a text input. The more exact the prompt is, the closer you are likely to be to your own ideas. For this, however, one must first know what the AI actually understands by which text. Not only is it important which term is included in the prompt to describe it, but also which arguments are added.
Fortunately there are tireless community members like Will Wulfken, who have invested hours, maybe days, to execute the basically same text input with different arguments and thus provide a clue to understand how Midjourney thinks. He doesn’t just limit himself to the current generation text to image engine, but compares version 1, version 2 and version 3. Unbelievable!
The result is a huge, constantly growing database with now more than 2,500 entries and more than 7,000 images, which can be viewed in its entirety on Github. If you want to thank Will personally or join the discussion, you can find the Discord thread here.