Version 3.0!!
What does this means?
-- I have overhauled this LoRA so hopefully it will work, and look, better than ever before! Please let me know if theres anything that needs to be fixed and i will look into it!
Since this was my top LoRA i thought i would update it for you all who like it so much! PLEASE ENJOY AND SHOW ME YOUR CREATIONS! THANK YOU! I LOVE EVERY ONE OF YOU!!!
Hayase Nagatoro from Don't Toy With Me, Miss Nagatoro!
Good weight is between 0.55 ~ 0.8
Have fun if you use it!
Description
Added a TON of images to the dataset and changed a couple of training parameters around so it has a better quality
FAQ
Comments (17)
Nice work.
I'm curious what do you recommend to train a lora.
I'm having some troubles sometimes, any advices that you could give me perhaps?
Yeah i can try. My advice would be to have a very good dataset. The dataset shouldn't have anything other than the person or style you are trying to get. Don't have anything blurry or out of focus either. Anything between 30 - 100 photos are good. You can have more but between those two numbers are good enough. For training i would say to use a checkpoint that you think will work well with your LoRA. Its all personal preferences but it helps to think that " oh i want to have it look more realistic so i'll use stable diffusion 1.5 base checkpoint." or "i want to have a more cartoon\anime look so ill use NAI checkpoint." things like that. For a more indepth guide you can look at these links here. It goes into great detail on how to train with ease and it helps me out even with this LoRA i made. https://civitai.com/models/22530/guide-make-your-own-loras-easy-and-free. Heres another for training locally: https://civitai.com/articles/1824/how-to-make-a-lora. These have great advice that i follow as well. Hope this helps!
@Vaporvvave Thanks! they are good advices i would try that! for a 100 pics How many steps/epochs do you think are recommended?
@Hikarias 10 epochs works really well with 100 photos
@Vaporvvave Many thanks :)
@Hikarias No problem! Glad i could help :)
v3 gives amazing results! Thank you!
Version 3 just perfect! I like these Nagatoro))
I am not getting these kind of results. The images don't look as polished and sometimes the model barely resembles nagatoro. Eyes seem to get all messy as well. Any suggestions?
Are you using stable diffusion or the civitai creator? If it's stable then do you have hi resfix on?
@Vaporvvave Yes i am using stable. I just do not know if i need any other things. such as specific VAE or like you stated hi resfix. not sure what that is.
@joshnome87523 Oh yea i can point you in the right direction. There's a lot of info here so i'm going to link you a video that helped me out when i first started learning about sd. Here's the video: https://www.youtube.com/watch?v=iAhqMzgiHVw&t=1s and if you don't understand anything then i am more than willing to try to help out! There are also good articles on tutorials on stable diffusion on here as well! As for VAE its more of personal preference but i can give you two that i use to start you off! A lil insight is that VAE basically helps out with the color of the photo.
Here is the link for the animevae, which is a new one that i am trying out now: https://huggingface.co/a1079602570/animefull-final-pruned/blob/main/animevae.pt
Here is the link for kl-f8-anime VAE that i use for the best overall color: https://huggingface.co/Norisuke193/kl-f8-anime2/blob/main/kl-f8-anime2.vae.pt
@Vaporvvave Awesome! I will be sure to check this out once i’m home. Thank you for the insight and info. I’ll most likely take you up on the offer of extra assistance so I appreciate that.
@joshnome87523 no problem. i'm always happy to help out!
Hi! Do you mind sharing the training data? This one works great and I've been trying to practice my Loras by referencing well made ones such as this :)
Can you do senpai?
maybe. I wasn't too invested in making him, but i might give it a try after I'm done with other projects.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.