A LoRa that aims to imitate Noise's artstyle.Again, en emphasis on aims here, because as you might notice I'm on version #4 here because I'm experimenting with training settings to get the subtleties right - and I'm still not 100% satisfied. If it looks too generic, you could try prompting lower scores, increase the weight or fiddle around with the negative prompt (less "quality").Trigger is Noise style.As with always, uncensored base data is scarce, so prompt against censorship for the NSFW stuff.

Noise artstyle (Doujin, Pixiv, etc) [Pony]

Experimenting with batch sizes and network alphas here. Best result - so far.

Files with SHA256: e7e35cd1ee8dd075eee0b3397812a70cea7a19965864562d1c56b6d898396b6b