Mayuko Yukimura from the manga series Katekin
Thought I'd try testing a LoRA with full manual tagging and some different settings. It's mostly trained on manga images but there are some colour images in the dataset so it can recognise the right colours. I'm not expecting anything amazing from this. It was more a test to see if I could train something with few tags. She has a gym outfit and swimsuit but mostly only appears in her school uniform or some variation of it
If using the jacket remove the short sleeves
(TAGS)
Character - MayukoYukimara
Character - long hair
Character/Optional - twin braids (remove long hair)
Optional - glasses
Outfit - school uniform, shirt, short sleeves, bowtie, skirt, plaid skirt, socks
Optional - jacket (As mentioned, remove short sleeves if using this)
Character - monochrome, greyscale (It's an 80 image dataset where I think about 76 of those images are black and white, this is probably going to be needed)
Bust size - large breasts (She seems to consistently appear with this in the dataset)
(TAGS to add/alter for Coloured version)
+ black hair
shirt > white shirt
bowtie > red bowtie,
skirt > red skirt
jacket > black jacket
Description
Trained on 80 images
Check description for info on adding colour but it was almost entirely trained on black and white manga images
Comments (3)
Not gonna lie this V1 version will probably be bad. It was a test to see if I could manually tag only what was needed plus anything which blatantly obvious in the dataset
If I get time I'll probably rerun the dataset through the good ol' 0.35 autotagger and try retraining it again
Also in hindsight I probably should have tagged her lips. Didn't realise how dominant those would be in the training by pruning them
Ngl, I thought you were working on your most popular lora. When will you schedule that?
@ColabThree2 I have been working on Eureka on and off however it is not an easy LoRa to train. I don't also don't get paid or take money for LoRA making so I do it in my free time when I'm not working and can motivate myself to do so and on top of that when I do work on the LoRA by training a test version I'm doing that locally which means my computer is basically out of action until it's done and if I play a game to occupy the time while it's training it risks the GPU running out of memory and the whole training crashing
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.