This version is trained on images of illustration, it has a strong imprinted style. Also it's not stable, so don't expect too much :p










Today i am gonna share with you, my Asagi lora.
It is not perfect, it's unstable and it might need a retrain in future. That said, let's talk about this lora.
It's true that there are few out there, but even toh i think they work pretty good, they feel so limited, i like flexible models so i've tried to train one myself.
It produce images of asagi similar to the original asagi of kagami irotaka
Yes and no, this model is more flexible than others, but it's also a lot more unstable.
The images i've uploaded as example are self explanatory, it can be used to produce decent outputs, but it need few attempts to do that, and some tweak (case by case) of lora strenght.
If you want a specific pose, concept or scene, it's easier to get it right with this one compared to others.
It can be activated using:
asagiigawain the positive prompt
This tag doesn't need further tags for hair lenght and hair color, but might need the tag "green eyes" to get the proper eye color.
This model works best between 0.6 and 1.
At strenght:1 on some models it tend to generate images with blue backlight/glowing.
To kinda fix this you can put the second tag
bykagamiirotakain the negative prompt, but doesn't work consistently.
If you want to produce a good image, i suggest to use highres fix. upscaling x2, using anime sharp, with 10 steps and denoising strenght at around 0.15.
Adetailer can be used aswell.
The model works best at 512768, 768512, 768768, 1024768, 7681024.
It has an idea of what a taimanin suit is: you can achieve a decent suit by using the following tags:
, purple_, _shoulders, gloves, fishnets, fishnet_gloves, impossible_clothes (this tag is optional)The size is variable, but if not defined it will tends to generate huge/.
It has a somewhat high lora compatibility (including style loras).
Give it a try, but don't expect too much, I aknowledge that this lora could've been much better.
This version is trained on images of illustration, it has a strong imprinted style. Also it's not stable, so don't expect too much :p
1. Yeniden paylaşılan modellerin hakları orijinal yaratıcılara aittir.
2. Orijinal model yaratıcıları, modellerini talep etmek için resmi kanallar aracılığıyla SeaArt AI personeliyle iletişime geçmelidir. Talep etmek için tıklayın
