Please, check the description, there is a little bit of useful information.
Indroduction:
This LoRA is supposed to represent Rabbit's artstyle.
Be aware! Try not to use any "quality" embeddings with style LoRAs.
Should work good with most character or concept LoRAs.
Artist's main features:
Cute anime-styled faces
Soft silky skin
Thick and long eyelashes
Puffy nipples and areolae
Recommendations:
1.0 emphasis.
Anything VAE is more close to original colors.
Precise prompting for nsfw generations.
For example - adding "nipples" or "pussy" tag to "nude" in prompt
Don't use too high emphasis values on tags.
Don't use "quality" embeddings with any Style LoRAs. Embeddings are chaging style too.
If you want to use "covered nipples" tag, be sure you've put "see-through" in negs.
If you want the same eyes quality use Adetailer or UDetailer extensions.
If you're ran into any problems with this LoRA, you can always dm me on discord:
troubledarkness - discord nickaname
I'll be glad to help | Буду рад помочь.
Credits:
Dataset of 150 pictures (v1), 20 epochs, Clip Skip - 2
Dataset of 165 pictures (v2), 20 epochs, Clip Skip - 2
Really big thanks to:
rabbit (wlsdnjs950) (erabi_) for beautiful artwork
Some good folks from Discord server
You, for reading, downloading and reviewing this LoRA
Description
Initial Release.
FAQ
Comments (4)
Hands down one of my favorite Loras to date.
The same. My favourite style right now. Works great for any character type, breast size or anything else.
From your own experience, can You explain specifically, what can cause an artist style to be very hard or impossible to train in a model? How do You notice when an artist can provide a good dataset to avoid under/overtraining?
Also have you tried to train a model with an art style that comes only from mangas/Doujins of a specific artist? Is it possible? how it could be done?
1. If your artstyle is too close to model style (that you're training on), it can be very hard to train style LoRA, it just can't find difference between trained style and your model. If you want to avoid this, you really need to use NAI for training any ->anime<- LoRA.
2. Most artists with at least 100 good pictures for dataset can be trained well. For best results it's better to have at least 300 (+-50) images.
3. I've never tried to train any manga or doujinshi style, but it would be harder because of a lot of different frames on the same image. And of course, there is more chance, that your autotagger would make a mistake. So, you need to check all captions precisely.
Details
Files
Rabbit (wlsdnjs950).safetensors
Mirrors
Rabbit-wlsdnjs950-style.safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950) (2).safetensors
Rabbit (wlsdnjs950) - style.safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950).safetensors
Rabbit (wlsdnjs950).safetensors
Style - Rabbit (wlsdnjs950).safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.











