At the moment, I’m sold on Midjourney. Although the learning curve is steep, the software does a bunch of cool things to generate images. It incorporates deep dreaming, which involves enhancing and modifying images based on patterns learned by a deep neural network. Say what? Here’s an example. Think about looking up at clouds in the sky and seeing shapes that look like animals. Your mind has created the images from the light and dark patterns in the clouds. Deep dreaming is the same concept, only a zillion times more powerful, and computer driven.
Midjourney also incorporates style transfer, which merges the style of one image with the content of another. For example, I can ask for an image of a roller coaster in the style of Van Gogh’s “Starry Night”. I don’t use this specific function in my own work, but I do use similar capabilities in creating an overall feeling once I’ve placed the main characters.
I also like the fact that it’s possible to tweak the results over and over, perhaps changing the word “large” to “enormous”, or “blue” to “light turquoise”. Sometimes the results are surprising – in scenes with humans and animals at least one rendering will have the animal’s head on the person’s body. I never ask for that, but often it’s quirky and cute.
I could go on, but until something better comes along (I’m know to be fickle, and I check out everything that’s new), Midjourney is my AI engine of choice.