Lora model for Katou Megumi from Saekano. I used it with weight 0.8.
I trained this cute girl in a different way but seems to mostly work fine ^_^ I noticed it's harder now on the website to include whole set of words in model parameters so I'll just be listing them in description more clearly.
Main prompt - katou megumi
Ponytail hairstyle - ponytail
Iconic outdoors outfit - white dress, red jacket, neck ribbon, beret, white thighhighs, frills, bow, mary janes. If you are a man ignorant to female fashion like me, mary janes are the shoes.
School uniform outfit - school uniform, white sailor collar, pleated skirt, kneehighs, loafers. I messed up training here a bit and about half the time you will get white uniform >_<
Bikini outfit - striped bikini, optionally sunglasses, eyewear on head.
Description
FAQ
Comments (5)
this is way too overbaked and doesnt work for me
Sorry to hear that :( What model are you using? Maybe I can give some advice on how to get better results with whatever you're trying to do :)
Edit: saw you added some pics but without metadata so I guess my input is not wanted ._.
@Deto15 site is broken so cant add metadata. it also somehow removed my comments.but yeah i will give you my input and ask for your opinion. All of your models i've tried so far seem overbaked to me. Which means with minimal prompt input they look good. But the more prompts you add or if you try to combine it with any other Lora, they completely break. Seeing how there is not many good example pictures people posted i will assume many others had similar issues.Tried it on some grapefruit and aom3 mixes but still coulnt get good flexibility with prompts. First Example picture i posted for instance was supposed to be a full body shot but it turned into a closeup of her face instead
@Wizi Good to hear back from you :) Yes, the website gets new bugs by the day, doesn't it.
I suppose you're not using Automatic1111 GUI? Metadata gets embedded into png files there so it's much easier to troubleshoot that way. I think even when website works fine it doesn't let you to input all the data manually.
You must have saw that 4th image in my previous is a full body. I feel prompt there isn't that short and weight of Lora is 0.8 so at least not as low as 0.5. I just now tried to run the same settings (except for seed) and generated 10 different pics on grapefruit, hassaku, cetusmix, kotosmix and anything 4.5. Every one of them ended up with full body (no, actually one had head out of frame ^_^;). Not to mention that if the fact that you got a face close-up instead of full body was a sign of overtraining, wouldn't that happen if I had close-up images in the training set? Because there weren't any - upper body at best.
I don't want to disregard what you write and would feel better if I could get to the bottom of this, but I hope you can see with all that it just makes much more sense for me to assume you might be doing something wrong? Especially that picture with embedded metadata would explain more.
I know that especially my older models are not perfect, which is why I don't advise weight 1, but you shouldn't have to go as far as 0.5. I think they can't be that bad... except for my Envy one which I'm currently redoing.
@Deto15 i am not saying this cant generate full body. i am saying it doesnt mix well with other loras or complex prompts. Did you try generating any images mixed with any other cloth or style lora, or concept lora. If you did successfully please upload. Because so far i tried 4 of your LORAs and none of them work well with other loras or highly descripted scenarios. If your prompt is "1girl,standing" these loras work well.
Also i mostly used your recommended weights. But ye would appreciate if you tried your loras with other loras and upload, i will try to upload one with metadata as well.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.