Eternal_Explosion

Hello everyone!

I'm so glad to present you my new merge today! It's actually pretty funny, because it was quite obvious that I absolutely should try it, but somehow I wasn't trying...

Anyway. Today we have an Eternal_Explosion, a nice mix of my two other checkpoints: Cyan_Cloud and Dark_Demon. At first it was intended to inherit DD's scenery and ability to interpret stupid prompts, while also being more clean with details like CC.

Buuuut... It's not like that. Yeah, it got DD's ability to read prompts and its intentions to build complex scenes. But colors? Oh damn, it was even enhanced somehow. New merge didn't become much cleaner (yet it did a little), but it's even more mysterious and stubborn :D

I strongly recommend to try it if you like non-ordinary solutions — it's definitely one of them.

So, a little bit of yapping about prompting.

1. I think you want to use masterpiece, best quality, newest like a no-brainer prefix, all the time. You may ignore it, but tests have shown that you probably don't want to.

2. You may use artists tags, model is totally fine with it. I think. At least my results using them were quite satisfactory.

3. If you want the madness — throw some weird tag-like nonsense into it. Something like psychedelic crystal cave, reflections, dark luxury cursed silk, kaleidoscope right after quality and character tags. Just some weird words and phrases, as if they were tags. PROBABLYYY it's better to just use normal booru tags — I can't say. I feel like it works better in a chaotic weird way, but I totally may be mistaken. Try your own way, for sure you may outbeat me with it!

Little extra tip: styles like vaporwave and synthwave? GODDAMN, ABSOLUTELY! Don't forget your digital glitch, dude!

For negatives — I just always use these:
(bad quality:1.1), (worst quality:1.1), bad details, (bad hands:1.2), child, loli, pubic hair. Sometimes I also add letters, shiny skin — when I want to get rid of, well, extra letters or shiny skin XD

About generation? Nothing too wild here, everything as always. Here is a quick explanation how I do it.

First Pass (Base Generation):

• CFG Scale: 5.

• Sampler/Scheduler: I use res_3m and beta57, but standard Euler with normal or beta should also work fine if you prefer simplicity.

Upscaling (I use Latent Iterative):

• CFG Scale: 4.

• Sampler/Scheduler: dpmpp_2m with beta worked well for me. I haven't tested other combinations yet.

• Steps: 4. It's a total overkill, 2 will do just fine, I just crave for maximum details.

Honorable mention!
My dear dude @haiti5 made a style LoRA for Anima models, based on my images. So, if you want to use Anima, and for whatever reason want to get that same aesthetic - please check it out! It'll be in the suggested resources below.

After all being said — please use some of the advice above and I hope you will get something beautiful! I'll be happy to see what you guys can create with this new little treasure (that I'll absolutely overuse for next 2k generations, I promise :D ).

Please, have fun. Have a good day!