okay so don't use v6 keep using v5

Also please don't generate children

ur sick if ur generating children bro!

@ilikeperc Please, only those with a mind full of pornography would immediately associate the image of a "child" with something inappropriate for children. You're saying it as if there isn't a single young character in the danbooru database

@SEU_Limbo From the bottom of my heart I hope it's not your case but you can't blame anyone for assuming, specially in the AI scene. Take a look at the "images" section, 90% of generarions are straight up porn and the rest are probably kind of suggestive at base. Also this model, tho I must admit I use it effectively a lot entirely for non-nsfw images, is vastly made for +18 shit too. In general, unless you want to make some Made in Abyss fan content, I generally don't see why anyone without pedo shit on their brain would want to generate a child, very much less so with this model in particular. Not accusing you of anything but you can't blame anyone for assuming stuff after reading "room for improvement in generating children characters" imo.

@Cee_One As everyone knows, the most important evaluation criteria for text-to-image models are consistency, generalization, and robustness. A model named after an art style should logically ensure good consistency across all elements within that style. The reason I make such a statement is simple: any slightly professional practitioner in this field who truly holds high hopes for this model will definitely find a sufficiently comprehensive test set to test it, especially character datasets. This is because the place where text-to-image models truly create immense value is not in the images generated by us here according to personal preferences, but in industries such as genuine poster design and advertising design. In these application scenarios, you cannot expect to never encounter a situation where you need to draw a child. If the model performs poorly in this regard, it is necessary to point it out in the comment section so that general users are aware of this to avoid wasting download time. Only if my statement was "this model performs poorly at generating pornographic content" would I readily accept your accusation—because the aforementioned application scenarios would not use such images. Beyond that, the so-called logic of "most people are messing around, so don't blame me for preemptively criticizing you" is absurd. Even setting aside your conflation of the model's capabilities with my personal morality, isn't the legal principle of "innocent until proven guilty" adopted where you live?

@SEU_Limbo Buddy this model explicitly states "（Sales are prohibited. Please use for personal enjoyment）" so don't come sell the industry use argument here. It would've been more dignified to just say "I'm not doing lolis bro chill" instead of this long block of bullshit lmao. Anyways you do you, as long as you ain't doing cp I honestly don't give a shit about your life at all, good riddance.

@Cee_One  Just because the creator here wrote it means nothing. Licensing is tricky.

Comment by someone on one of their models :

"Commercial use of the model is authorized under Illustrious' standard license and cannot be modified. Anyone can download this model and do whatever they want with the outputs (not just this model, but literally any model using the Fair AI Public License 1.0-SD).

Download, generate images, use in projects, sell, do whatever you want"

I genuinely 100% don't mean to self plug but the LoRa adaptation problem is only ever there on models that were trained with a starkly different checkpoint, LoRas made with this checkpoint as base work great if you are willing to look for them or train your own! All the models and LoRas I've done were trained with this checkpoint because it's the one I like the most and want to use when generating :)

For the finger/hand thing I reccomend ADetailer for the yolohand and the "lazyhand" textual inversion from the LazyEmbeddings pack, two real lifesavers.

Hope you get your results to be a bit better with this, cheers!

@Cee_One Hey hey!!! I'll take a look on your LoRas!!! Thank you for advice and good luck in your LoRa building!!!! :)