Description
FAQ
Comments (12)
Trigger words and recommended weight?
Also, is this for inpainting?
tested on 0.6-1.0, that’s fine for me, no trigger word needed. Appropriate for your test and share your results, thx!
Nice LORA! Do you use trained images more than 20? Did you also tag eyes during training?
Thank you! Exactly 20 I used, tagged ‘perfect eyes’ :-D
@ChotomaTo Cool, using only 20 training images is art itself
Anyone else having issues with this washing out the colors?
Based on other review images, the lora model doesn't seem to be what's doing that. Try using a vae or switching out your existing vae in the settings of whichever UI you're using to run Stable Diffusion. There are a few to choose from, but off the top of my head you can try ft-mse-840000, kl-f8 anime2, and clear vae. You can find most or all of them here on civitai too
@Bunger_Beesechurger idk... im using this in combination with the anylora checkpoint and i tried three of the anime VAE's and it doesn't really change anything. I am using this with a couple other loras too but when i adjust the weight of this lora it seems to make the hair pinkish instead of red.
@ericaapplespeanuts239 Still, no review image for the lora has washed out colors, which is the weird part, and without images or settings to check, I can't really say for sure. A lot of vae's are actually the same file renamed and repackaged with popular models, so that might be the culprit. I guess unless you're dead set on using this lora (which is a throwaway/test model with no description anyways), I'd probably just move on. I was told by multiple people at different times that it was only the base model and vae that determined if an image was washed out or not. Guess it's time for me to ask them again and see what they say
@Bunger_Beesechurger thanks for your help! I appreciate it. I really like this model, but if I were to set this to 0.8 or higher it starts messing with the colors and I tried a couple different checkpoints and its does the same thing. i like the subtle details and lines it adds to the image and i can't seem to replicate a similar style with some other loras that draws outlines. i guess my only solution right now is to set it to 0.7 or lower
Issue seems to be they are all blue. Not sure if you've seen us asians before (Filipino here) but there needs to be way more eye colors included in this training.
I don't think the lora is supposed to be for Asian eyes at all. Not sure where you got asian from... It seems to be a throwaway model for the creator/author's preference of a certain type/look for blue anime eyes. They just posted it in case anyone else wanted to try their test
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.