This lora is all about looking cute, dainty, and also chaotic~ dataset from Niji v6.

Only downside? It prefers longer prompts + using the trigger words since it wouldn't work if you simply put random words. Also since the dataset is about certain lightings like soft, chiaroscuro, etc, the eyes might BLEED like it might show up sore eyes in some certain way.

Though it's pro side includes: surprisingly good fingers, dynamic poses, can do good cluttered stuff, hard styles like Gyaru and decora, and can do super deep shades of skin tone such as ebony. I've been trying so hard to figure how to force tensor for deeper shades since I really love how ebony looks. Also for whatever reason ebony x pink is so damn gorgeous here, especially with those dollete and coquette style, I'm going to die out of cuteness. I really cooked with black women dataset. Ironically white women gens looks like anime if u put it simple prompts in prompting lol good luck. At least the Asian gens looks fine, might need a bit backup🤣

Recommended sampling steps: 8. You can do higher if you want more dreamier atmosphere, sharper details. I prefer 8 since it's cheaper honestly!

Strength: 1 or 1.2 as a base. As a support idk up to you, 0.3-0.8 depends. It's a good support lora for heavy semi realistic loras. It's krea so the style doesn't go wonky unlike Zit or illustrious.

Trigger words: nijijelly, polished digital illustration, pseudo-realistic portrait art style.

If you don't put the trigger it will look like anime/comics!

Models: I prefer using rRealisMAX Krea-2-V1.5-aio since Krea 2 BF16-TURBO and Krea 2 TURBO has this oil painting vibe. The BF16 version is better though, it has this more jelly vibe.

Version sweet and sweeter are more closer to anime/comic so I named them those🤣