Here we have Alicia, a pretty oneesan of one of my favorite slice of Life shows(the other being Yokohama kaidashi kikou). Let's be honest here... who else thinks Akatsuki got Alicia stolen under his nose by a rich ugly client? That's my head canon. I don't mind it either as i don't like Akatsuki very much. So this one has my Netori seal of approval I guess(just like any SAO in which Kirito gets Asuna stolen. Fuck Kirito).
This was trained mostly using the manga, some official art and a nice doujin by circle outerworld and a couple of others. The color was mostly provided by anime screencaps. I tried to be as congruent as possible as Alicia looks quite derpy in the manga from time to time. I did my best to try to retain her more serious face.
Seems to work fine with: <lora:florence_aliciaV1:0.7> florence_alicia,
I isolated the following outfits/parts of it:
Summer_aria_company_uniform_White_long_dress_White_capelet_Blue_bowtie_Short_sleeves_Side_slit: The summer version of her uniform.
Winter_aria_company_uniform_White_long_dress_White_capelet_Blue_bowtie_Long_sleeves_high_collar_Side_slit: Winter variation of the aria company uniform.
White_beret_aria_company_uniform: Her aria company uniform beret.
white_tankini_bikini: Her sports bikini from a beach episode, she also wear her hair in twin buns.
bespectacled_Round_eyewear: Just her using her glasses, she modifies her hairstyle a bit when doing so.
I used too much of the manga and anime screencaps so no dataset for you this time. Instead I will only add the tag summary so people can experiment.
Description
FAQ
Comments (5)
I am not sure what happened but the Lora file wasn't uploaded, It should be fine now I think. So if you got a zip file it is only the tag summary.
It depends on several factors like the optimizer and the amount of repetitions, but normally when using adamW it was fine with both at .0001 and about 8000 total steps per concept. Nowdays I stick to prodigy with a lr=1 with about 4000 total steps per concept. This is my guide: https://civitai.com/articles/138/making-a-lora-is-like-baking-a-cake
@knxo thanks, i am traing lora= Colab
@long04091988192 I am not really sure how to configure the training in collab(as i train local) but most collabs are kohya based anyway so the settings should be the same if labeled a bit differently(easy training scripts runs kohya under the hood).
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.











