@Ikena ok, thx :D, and one more question.. do you know many zzz characters are trained in this ckeckpoint? i know maybe i should try for myself... thank you for helping

@Ikena is there any way to know which anime character are in your checkpoint?

@Ikena I try trigger\(zenless zone zero\), but I get the wrong result.

put this in your negative prompt:
chromatic_aberation_(color-burn_(azure):.15)

I noticed that too. The model tends to make the image look as if it’s in cooler tones. I compared it with versions 2.1 and 2.2, and for example, in beach scenes — even when the prompt specifies that it’s a sunset, which implies warmer lighting — in version 3.2, many images lean toward neutral tones, as if it’s midday rather than sunset.

what

@crombobular Just kidding !, No sense of humor, he!, meant for multiple character complex posing.

@Ikena I’m using something like this:

 retro, retro_arstyle, score_9, score_8_up, score_7_up, ultra detailed, higly detailed, masterpiece, best quality, very aesthetic, anime screencap, masterpiece, best quality, stunningly artwork, best quality, stunningly artwork, anime style, studio anime, source_ecchi, ultra-detailed, best quality, amazing quality, very aesthetic, high angle shot, top view, nsfw BREAK

1woman, solo, detailed eyes, glossy pink lips, detailed eyes, detailed face, perfect body, perfect face, sexy, flirt, seductive, micro skirt, lace thong, full lips:1.2, knee-high socks, sitting on professor's desk spreading legs, soft lighting, sensual atmosphere, highres skin texture, open neckline, seductive BREAK

I tried turning LoRAs on and off, but for some reason the model keeps making the teeth sharp by default. The only thing that helps is specifying in the prompt that the lips should be closed. Before this, I used version 2.1, and there were no issues with the same prompt and loras. I don’t know what’s causing it.

@mrf1sheragent783 I would do:
1girl, mature female, long sleeve shirt, partially unbuttoned, cleavage, collarbone, miniskirt, lace thong, pink lips, thick lips, [pick an eye color], long eyelashes, kneehighs, sitting, on desk, legs apart, classroom,
You don't need all of those quality tags. Those are more useful in Pony, not Illustrious. Use danbooru tags to get the best results, and good luck!

@Ikena No worries. Yes, I managed to solve the problem. And indeed, reducing the number of quality prompts to a minimum helped a lot. I also noticed that version 3.2 naturally produces more detailed images, so LoRAs related to enhancing detail should be given lower strength or even disabled entirely. That way, the results turn out much better.

Even though I like versions 2.1 and 2.2 more in terms of style, it’s clear that 3.2 produces better images by default compared to the base 2.1 and 2.2.

Thanks for your reply and your help! By the way, I want to say that your models are among my absolute favorites. I create about 80% of all my art using Hassaku. Thank you for your hard work — keep it up, and best of luck

@tesylmu Thanks for the advice! As I mentioned to Ikena, yes — I found that in the case of version 3.2, reducing the number of quality prompts and lowering the strength of LoRAs related to detail enhancement helps achieve a better result. Version 3.2, even without any LoRAs, produces more detailed art by default compared to previous versions. So overusing LoRAs and prompts in this case heavily affects the final result. Thanks for not being indifferent to the issue!

@mrf1sheragent783 Each base model has its quality tags. "score_9, score_8_up, score_7_up" is Pony. It has nothing to do with Illustrious. It's not booru tags at all. Don't get used, like too many people that can't think, to drag tags from model to model just because you're used to copy paste them. Score_X means: nothing to IL.

Your prompt is super long for no reason as most of it means nothing to IL (invalid tags, or 0 entry tags in booru)

SDXL, Pony, IL / Noobai = different way of prompting.

Use Danbooru webiste to check a bit all words and get cleaner prompts.

You'll need an extension like Forge Couple to do so with any consistency.

@mostlyforhentai is there an option for comfy?

@Fgonax Yes! https://github.com/Danand/ComfyUI-ComfyCouple

@Ikena how do I prompt it? Everytime I prompt an artist it just uses the base model style