NSFW style: Bai Asuka
Slightly smaller data set and rushed training compared to my other nsfw style loras so might not be as accurate at style reproduction. It is also my first try with a non-trivial amount manga panels and covers trained in, so might wanna put things like “text, cover, comic, comic panel, monochrome, etc” in the negative if it bleeds.
Main trigger is: baiasukastyle, (and “uncensored,” in the positive plus “censored, mosaic censor, etc in the negative if you want to use it for the horny)
Hope you like it!
Description
NSFW style: Bai Asuka
Slightly smaller data set and rushed training compared to my other nsfw style loras so might not be as accurate at style reproduction. It is also my first try with a non-trivial amount manga panels and covers trained in, so might wanna put things like “text, cover, comic, comic panel, monochrome, etc” in the negative if it bleeds.
Main trigger is: baiasukastyle, (and “uncensored,” in the positive plus “censored, mosaic censor, etc in the negative if you want to use it for the horny)
Hope you like it!
FAQ
Comments (2)
Yeah I added some material from what seems to have been earlier works as I was struggling to find enough colored images. They were all tagged with the proper artist, but comparing them to the manga you posted as examples they are different. Only noticed how different they are after about halfway into model training as I was rushing to start it so it finished in time and didn't have enough time before deadline to retry from scratch before bounty expired. Was hoping they would not overwhelm the style completely and can be prompt controlled with a 2/1 ratio but I guess they did overwhelm it. Only confirmed it after posting with more testing >.<
Glad you still liked it! Honestly I am a little happy that the merger of the old and new styles resulted in something unique that is likeable rather than just a mess ^^
I'll probably rename this version "old vn style" for that reason, and maybe come back and try to do a v2 with just the newer images at some point once I get through the backlog of character loras i have waiting for training lol
Do note I am not sure how well it will work since the dataset will be about 1/3 the size and much more manga panel heavy without the "old style" images though, and it will probably just be similar to what the other entry produces, assuming he only trained on the newer style dataset.
Lessons about rushing my work were learned this day xD
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.