This LORA is experimental as shit. Its cursed. but it's funny as hell. I hope you enjoy it. But if it gives you nightmares, thats on you. You've been warned!
I WANT TO SEE YOUR PICTURES THEY ARE ALL BEAUTIFUL I CAN TAKE IT
==Update for SDXL==
Big super awesome kudos to @Jabberwocky207 for showing me the true potential of this portal into hell - I love you
Check examples, works well with TurboDiffusionXL. It's prompting is a little different, more flexible, but less fucking batshit crazy too. Sort of.
===vvvv SD1.5 vvvv==
I worked with a guy 20 years ago that had this on his desk, so it's not just a meme, they exist.
It works pretty good with Sinistrella, and works more realistically with Epicphotogasm. I'm sure there are others. This shits on anime models, but if you find one that doesn't make it look awful, let me know or post pictures.
Eular A for a more unhinged take, works better with double exposure, but is awful at cats and faces.
DPM++ SDE (not Karras) was having the most luck, as it's a little more dreamy, and euler A is more likely to actually do the double exposure better, but is awful at cats.
That being said, poke around with the settings,
Go with basic negs, or go crazy with your neg soup, it seems to matter, but you'll have to find your own path there.
Try full strength 1
((double exposure)) at the very front, going wide res helps a lot with cowboy shot
meme, woman or man
Trained on space background, laser background, black background.
cat, holding, holding cat, ,(photo insert)
Description
FAQ
Comments (8)
"the model is cursed" O_O
Yeah, don't taunt it, it takes it personally.
I like how we are at a point where we make loras to recreate bad photoshop edits on purpose.
lol I have No idea how all these lora models work, Civit-ai is confusing me a lot, I´m more the Midjourney guy haha :D
But here´s my meme of the day and I´m really not pro trump, I just wanted to test out what it can do =D
and why don´t I have these laserlights in the background? I just don´t get it.. sorry =D
If you look at my prompts, it's "Laser background" it's specifically trained as 40% of the dataset or so had laser backgrounds, other other 40% was "space background". I haven't had any issues getting them to come out when prompted. But definitely let me know if that still doesn't work!
This works shockingly well and is often hilarious.
Any plans on making a Flux version of this Lora?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.