Takedown requests can be submitted via Twitter at @AkioA_I
削除依頼はTwitter（@AkioA_I）まで提出可能です。
THIS IS ANIMA, NOT ILLUSTRIOUS OR NOOBAI! It is a different model and it won't work on those checkpoints/models!
Description
Trained on Anima Preview 3
FAQ
Comments (18)
Is film grain tagged on the images along with blurry? Also what res was this trained at? Been having a bit of trouble getting negatives to work. I like the newer style for the eyelashes and nails, but kinda dislike the film grain and depth of field. Thanks for the update.
Film grain makes it difficult to upscale.
@HaloSkull oh, well, I didn't tag film grain because most of Cass's images have a slight film grain on it. I think the old version also has film grain on it (not sure how much though compared to the newer one.)
But I did tag "blurry background" in the dataset or "blurry". The dataset hasn't changed besides me fixing the "multiple bows" situation. I added like a few new images into the dataset and that is it. If they release another Preview model, I will tag "film grain" on the images or will retrain and retag the dataset on Preview 3 when I have time.
It was also trained at 1024 like all of my other models.
@AkioAI Yeah it seemed like putting blurry in neg fixed that. Another issue I'm having with this version is it likes to put a white border in the background. I know a lot of Cass' latest imgs have like a small box of a background but it kept wanting to self insert without being tagged. I had to put border, white border, simple background, white background in the neg for it to behave. There should be enough of the latest style without the border thing for it to work still.
I've got 164 imgs of their latest style without the background border thing.
@HaloSkull Interesting. There is like only a few white boarder images in my dataset so I don't know why it would come out like that for you? I only have one single image that wasn't tagged properly that I have mislooked, but I doubt it would have effected the entire dataset because all of the other images have detailed backgrounds. Thanks for letting me. I will fix it next update.
It also has 100 images in my dataset as well. When I tested the lora, I didn't experienced that. Thanks for the feedback.
@AkioAI The Circlestone labs guy put out an official lora. He showed his training parameters.
@HaloSkull I will have a try of his settings with my next, next style lora. I am currently already training a lora with my current settings. I will take a proper look at his settings after. Thanks for telling me. 🫰
Thank you. Very nice!
It works great with many other loras as well.
Can you training the training settings?
amazing work as always!
Version 1.0 is out! Hope you can update with the few tweaks I suggested :D, Thanks.
It seems like it doesn't matter if I tag for "film grain" in my dataset (that is the prompt for noise grain) because the whole dataset has film grain. Cass has it in all of his images mostly.
@AkioAI Yes, but tagging that should allow for you to put it in the negative when genning to remove it for upscaling purposes (film grain and noise destroys upscaling). You'll just need to have it for the positive prompt when genning if you want the film grain and it'll look 1 to 1 with the normal style.
hope u update ur other style loras too for the newer version :3