I expect them to get symbols wrong for quite some time to come.
Looks a bit like a sheriff’s badge
I’ve had a few brief successes with a Star of David necklace prompt. Something Mid Journey always struggled with.
I also noticed that Stable Diffusion likes to ignore the full body (head to toe) prompts for character images. I usually end up with head to shoulders / ribcage character portraits. But these can have surprising quality.
2 Likes
The six-pointed sheriff’s badges look like the Star of David.
I am the Law (already)
Oh, just checked, the Soviet Star is red with a gold outline…oops (or phew)…
1 Like
lol
Anime is good for forgetting a butt on a male.
1 Like
I just installed the Fooocus * Stable Diffusion UI, experimented with a very basic prompt (Krista’s necklace, paraphrased) and a default setup, meh results. But this could be a useful entry-level tool for someone interested in SD. I’m more impressed by the stated system requirements. Fooocus could run on my old computer (in theory, with some luck).
*
1 Like
Openart is changing things up. Some things will only be behind a pay barrier, but I’m not seeing a ton that is, yet.
1 Like
More playing with Midjourney.
This time I’m recreating characters from my fanfic.
Unfortunately, creating children, I keep coming up with censure problems. Appealing sometimes works, sometimes doesn’t. I’m not sure which word combination does it since these were all very similar evolves. I’m convinced it’s the generation itself that was going to be lewd and MJ stopped it before showing it to me.
I’m always worried if behind the scenes they have a counter how many censures before they block me. That would majorly suck.
Jack and Elsa
Lots of good results with Jack and Elsa. They’re twins so they’re supposed to look alike so it’s not too hard to create them together but it took a lot of tries to get the right style, clothes, poses, facial expressions. The shebang.
They’re not wearing their Hogwarts uniforms here but it’s so good. I love it.
Individually, it was easier to dress them up and get the right age.
Merlin
Merlin prompts always come out looking good but it’s hard to get the right eye color even though I use Colin Morgan in the prompt and he has very blue eyes. Thankfully, that’s an easy fix.
I even tried to start with just the eyes and zoom out but results were disappointing. I’ll just stick to normal generation and recolor in Krita.
Update: almost the right eye color.
Also, I finally figured out why Merlin generations often have him staring into the camera so intensely. It’s because of the scarf in the prompt. I started noticing the annoying number of generations like this one where the scarf is on his head. I realized that they look familiar.
It’s that Pulitzer winner photo of the Afghan refugee. Omg. It keeps tripping me up, affecting every single generation of his.
I tried moving keywords around but the same keeps happening. Same startled pose and scarf on his head.
Serious prompt modification is needed.
Oh, I’ve also upgraded my cat profile pic. The new and improved KamiCat.
2 Likes
I can see it. Its a heavy influence.
It does result in cool generations. Like
Which I zoomed out into
But it gets in the way when I want him to assume a different pose.
So in an effort to introduce some variety, I tried to add magic and dragons. Results have so far have been tricky. I’ve noticed that when I evolve the image too many times, it loses the style and becomes sort of… plasticky. Not sure how to describe.
Pictures then
It took a lot of effort to get him onto a dragon-looking dragon, and by the time I did, the style was lost. Now it’s too cartoony and evolving it more makes it worse. I’ll have to start over.
This one is pretty cool though
Spell casting is easier for AI than dragons.
2 Likes
Working on Headers, while I edit.
Maverick, Nathaniel’s father…is an ass. It really came through on this pic, including the subtle middle finger.
I had to do some edits: adding lower lip, bulking out arms and legs, making sure there is ANY bulge where it belongs (I don’t know why that is pancake flat on models), lengthening middle finger on closer hand, cutting out AI green spots, evening beard and back of head hair (the mullet was only halfway expressed), smoothing random hair between eyebrows, color changing eyes, mirroring half the bowtie.
But the best part is the subtle middle finger came that way–I didn’t edit that in.
His sons are the header for the chapter before it:
So they all look vaguely alike enough for me to not go insane. That’s all I care.
But the hubby was cracking up because I was cackling as I edited “daddy”.
2 Likes
Possible theme music…
2 Likes
I tried to create a few flying characters from the NOOM. Results are awesome.
I have so many to choose from. Which ones? Okay, let’s pick a few.
Btw, I did not ask for abs or hotness or anything shirtless in my prompts. MJ just couldn’t help itself.
Warning, some of it is a bit NSFW
Creating Enlin has proven difficult. He’s a Sylph. Anyway, a couple of attempts that are not Enlin but quite the eye candy.
Tried to give him transparent wings. Shirt turned transparent instead. I’m not complaining.
Tried to create Reed, a garden fairy. He’s a lot hotter than I imagined.
Queen Pitsa. Not sure which one should be the official one.
Some fairy guards
And a couple of awesome flukes
1 Like
None of those are that indecent, I get occasional penetration over the word dominates. Like wtf…
1 Like
Update, he had me add a coin to Maverick’s hand.
An update to Automatic 1111 was released a few weeks ago, and it has some interesting features. My favourite improvement being better resolutions for up-scaled images. There is also a Ad-Detail extension, but I advise against it due to excessive pixelation around item / character edges (especially the eyes). Oh, I rarely use negative prompts now as most SDXL images are good enough without them. Image sharpness / clarity can also be a dice roll sometimes (regardless of prompts or Loras).
One downside is the up-scaled SDXL requires all of my Graphics Card’s memory / resources, to the point where even YouTube videos stall out. Apparently, the recommended GPU for high detail SDXL image generation is a 4090 24GB or above…If only I had a spare $3,500 AUS in my wallet…
3 Likes
This free generator is pretty solid. I mostly used the painted anime style.
