My dude and I were making some AI images, and while I’ve been doing more sensible things like the image on the left, he’s been doing glitchy weirdness like the image on the right. The one on the left is in Midjourney’s 5.2 engine, the one on the right is from a version created by and for Asian countries called Niji. I may have some terms wrong in all that.
I tried mashing them up by using the images themselves as prompts. Here’s just the images, in original flavor midjourney:
Here’s the same mashup, but also using the text prompt he used to get his original image set, “neon pixel art cover, in the style of horror manga, webcam photography, psychedelic manga, trick of the eye paintings, ascii art, disfigured forms”:
Here’s me switching from midj to niji, with just the images as prompt:
Then in niji with the text prompt included…
I think the winner is the last one, on the right. The one on the left has the head duplication problem from early midjourney outputs, like a few versions back. Kinda fun if you don’t recognize that, though. Nonetheless, cannot hold a candle to “lascivious psychic ghoul temporarily doffs the skull mask to use third eye on you.”