I was having similar issues recently, but managed to fix it by logging out, clearing the cookies and stored files in my browser for NightCafe, and logging back in again. The site had been basically unusable for days and this fixed it.
Although I haven’t posted on the SDMB in years, I sometimes lurk and noticed this thread when it was new. I don’t know whether to thank or curse @iiandyiiii and the rest of you, because I’ve since spent countless hours making AI art! I’ve appreciated the many tips and tricks you all have shared here.
Usually I never really try to craft an image outside of generalities; I usually just pull the slot machine lever and see what I get. I’ve been having fun evolving images over at Playground AI, notwithstanding. That said, I did my best to craft this one, though I’m not sure how recognizable it is (I saw some versions of this sort of thing on Twitter so I thought I’d give it a shot):
Kind of funny that an attempt at getting something weird (a mushroom girl, starting with a photo) ended up being one of the most realistic images I’ve generated. (The noise was apparently turned up too high.)
Anyone else playing around with Playground AI’s AI, Playground? You can get some interesting stuff out of it. I took the mushroom photo used above and did Playground img2imgs with a few prompt variations and got some of these.
I tweaked my favorite and created a wood paneling background for him (also using the Playground AI) and composited them together.
(ETA: I have no clue why embedded photos work in this post but not a couple of posts up. Same site, same formatting, different results.)
I’ve been experimenting with song lyrics as prompts on Midjourney. I take evocative passages with no other added descriptors, just lyrics, and see what happens.
Here is a verse from Bob Dylan’s I Want You, from the Blonde on Blonde album.
The guilty undertaker sighs
The lonesome organ grinder cries
The silver saxophones say I should refuse you
The cracked bells and washed-out horns
Blow into my face with scorn
Here’s a result from Playground scrubbed through a set of filters (basically v1 no filter, then evolve through 1.5 Delicate and then back through v1 Delicate):
Here’s another interesting result. I can’t remember the exact series of filters I put i through, but lately I’ve found polishing a piece up with a final scrub through v1 Delicate Detail usually produces interesting results.
I think most systems weigh early words in the prompt heavier than later words. Since the classic western undertaker is a guy in a black hat and suit, it starts with that as its basis for the image.
On the other hand, though an undertaker might be associated with corpses, he is not usually skeletal himself. The AI is (correctly, I think) interpreting the words via metaphorical associations, rather than literally. Which is impressive.
