if you want to support me click here
V1.5 dosn't need to follow this tutorial for good results (don't ask me about the training configuration cause i made this lora long time ago):
Suggest to read all for better results.
it's a 768 x 768 trained lora model (works generaly in every image size) of the hentai artist Cham22. this lora model is trained with clip skip:2, using the anything v4.5 model (don' t know if it run in any other anime style model) ¡if u have any problems with the colors of this model install the vae model and put it in the vae folder! while i was creating this model i added multiple tags that will trigger specific changes on the result of the image (some are just to change the style at some body parts, others are for the point of view, specific positions and etc.) because i'm to lazy for putting every tag on the tag list of this site i will upload a screen of every tag "added" (everything with "()" in the screen is for explaining some tags) with at the end an order i follewed during the writing tags fase that will surely help to get better results. generaly nudity activate the lora model
it's my first lora model that i upload so i don't know if this image is gonna be readable so i'm gonna upload it even in the exemple images. sorry for the bad gram and the long text.
Description
FAQ
Comments (7)
thanks for this. please train on NAI as the base in the future for better compatibility though since everything is derived from that
ok thx for the advice
How many images are in the training set?
334. Why?
@flusk I just think having more (and more varied) training samples helps the quality of a lora a lot. Especially with being powerful enough to produce the style or character without frying the image, even if there are a few other lora.
I don't trust sample images as much because they can be gamed many ways, even by accident. And running my own comparison takes time, so I'll do it later.
@worse-idea794 doing later what?
@flusk Comparing the quality of the Cham22 style loras. Some of them might affect the hair color of characters or their pose, which I don't want. Or one Cham22 lora is better at cooperating with other loras. One could be more accurate to Cham's style than the other. I won't know until I try them on the same models with the same settings and prompts (except trigger words if they need them).
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.









