LoRA is generated entirely from game images.
Should work well with any model.
Check samples for prompts :)
Had to make my own as I wasn't satisfied with results from other LoRAs.
The most reliable outfit is her black dress, maid outfit. Other outfits are a little bit harder to generate, try to use negatives.
Leave a like, comment, review but first of all, ENJOY ;)
If you wish to support me in my work,
you can do that here: Ko-Fi.com/LittleJelly
Description
FAQ
Comments (12)
i hope you will do the other NekoPara girls like Chocola, Azuki, Cinnamon, Maple, Coconut... ^^
Unfortunately, not anytime soon.
Unless someone provides me with game screenshots of them. :)
you could look on this sites for screenshots (they have multiple galleries that you have to look through)
https://exhentai.org/tag/group:neko+works
https://gelbooru.com/index.php?page=post&s=list&tags=sayori_%28neko_works%29+nekopara
https://danbooru.donmai.us/posts?page=2&tags=sayori_%28neko_works%29+nekopara
https://chan.sankakucomplex.com/?tags=sayori_%28neko_works%29+nekopara&commit=Search
but it can happen that you encounter a "sad panda" image, but you can avoid it, there are some tutorials or even browser-extensions for it
i can try to upload some of those galleries to my microsoft cloud and send you the link to provide you with the screenshots if you run into problems
@Enno89 nah, I know those sites well, but it's pain in the ass to find the official art there, not to mention there's a lot missing. E-hentai.org has galleries from the game, however it's time consuming to curate just one character. It took me 2-3 hours to curate vanilla (~150 images). That's the time I don't have right now.
Again, unless someone provides me with a dataset (no need for tags), I can't do it.
This is an example of curated dataset.
https://i.postimg.cc/y8bD0QjT/Screenshot-2023-07-23-104414.png
Curated = Filtered, cropped images so there's as little of noise(other charas) as it's possible.
Tagging images takes a lot of time too, yes I do use automatic tagger, however those tags need to be manually fixed if you want to have easy time when generating a specific outfit. Autotagger isn't perfect.
If you can prepare a dataset (no need for tags, i'll do that myself) as shown above for other characters,
I can run a LoRA training no problemo ;)
the model is very cool I heard you don't have a dataset for Nekopara, do you need a dataset :D
需要我为你提供数据集吗
I only had time to curate Vanilla dataset and I'm focusing on another anime right now, so I don't have time for other Nekopara characters unless someone provides me with a decent dataset :)
hahaha,I have a decent dataset, but it's huge and I don't know what route to use to transmit it to you
我觉得你可以在更新后的模型中添加正则化训练的眼睛 可以让眼部细节更加清晰
I don't speak Chinese, sorry :/
You can train your eyes using regularization in your training set, and you can replicate the original eyes almost perfectly
The models I trained in the past did not use regularization, resulting in blurry eyes, and now the new model uses regularization to achieve close to the original drawing of eyes