hello ai

+)Hmm. seems PIXAI fixed Animagine's non-turbo issue slightly. but maybe not. so i'm checking now. they should know, when the User stick to old model, even when the newest so-called "better-model" launched, there's a pretty solid reason. so maybe i'd rather keep use Animagine in here, until they remove it from training option. you know why? 'cause Tsubaki whatsoever and other DIT model's gen&LORA training cost is quite a insane. ◆When you use "(Anima)" LORA, i recommend you "Animagine XL V3.1" version by Cagliostro Research Lab in here. not 3.0, or 4.0, 4.1... others. it's strange, but it obviously give me a different result when i use some others. so as my experience, 3.1 is solid. and don't forget adjust your CFG to high level. (ex:min 8~15 max). it affects a lot, really. especially when you don't like "sepia tone&foggy detail" in the animagine's gen result. i know it's totally opposite for other models(like noob, otome, tsubaki...), but this animagine LORA, needs high CFG&high sampling value. (priority 1 is CFG. you can save your credit for testing at low level (ex:20~25) sampling, if you want. economize!) it's more affective when you needs manga, anime illust-like vivid color&character style. i used to input a at least 2000~3000 length text promft. hope it helps you. =========== well obviously there was certain "downgrade" update for "animagine" model's Non-turbo mode speed recently, when you use non-turbo mode for animagine mode, it takes a MIN 15HOURS~MAX CANCELD. AFTER 36HOURS OR WHATEVER. and PIXAI have NO intention to fix this problem. 'cause? they willingly want to do. on a purpose. so i'm quit to update my LORA training. yup, No more user contribution. No more tagging. No more reliable artist categorising. i'm done. who cares? they have already plenty of resources. and so am i.