详情
推薦
v1.0
Contemporary Pain

Contemporary Pain

18
29
146
#動畫
#女性
#風格
#女孩們

Denizens of this wastebin, I once again bring you a LORA of questionable quality, this time a blend of contemporary anime of questionable quality, which I recently (coincidentially) consumed like a pleb in manga form.

As always example images are provided with little tuning to provide a raw expression of what the model does.

Further coincidence includes the fact that for the first time not only did I dare to look at contemporary anime and manga, but also that all of the media includes ongoing or recently finished creations. Wohoo me.

The mix making up this dataset contains primarily Yofukashi no Uta, Gundam Thunderbolt, Goblin Slayer and Hellsing sprinkled in for fun, with a healthy drizzle of chainsaw man. I hear you asking: "But they look nothing alike! How could you even dare do this? I will drone strike your house!" The answer is that they look close enough for comfort (which is my primary complaint with contemporary animation, btw.)

My idea behind this was to use contemporary animé to get rid of that generic "AI" Look a lot of stuff has nowadays - while you can rid yourself of that relatively easily, I wanted an easier solution, as I am well aware that we are all a bunch of lazy folks. The solution was rather self evident. By using a dataset made up of contemporary productions, I could create something that would give a nice ammount of "real" flair, and hopefully reduce that "AI" feel of generative creation, effectively "muting" the combined feel of current models that are based on anime created in a certain time span, which inherently makes them suspect to spitting out results close to the time they were created, thus injecting a little bit of "believeablity" to the generated results.

If I am to say so myslef, I would dare to say, that I have succeeded pretty good!

I'd suggest weights around the 1-0.6 region (See the XYZ plot). Though the LORA seems to behave itself very well. (probably because of the adaptive algorithm I picked to try out), and as is probably self evident, this does best with anime models which lack a distinctive style of their own.

Use clip skip 2 you filthy animals.

查看译文

評分與評論

-- /5
0 個評分

尚未收到足夠的評分或評論

avatar
公告
2024-04-01
发布模型
2023-11-05
更新模型資訊
模型详情
類型
LORA
发布時間
2023-11-05
基础模型
SD 1.5
许可范围
來源: civitai

1. 轉载模型僅供學習與交流分享,其版權及最終解释權归原作者。

2. 模型原作者如需認領模型,請通過官方渠道联系海藝AI工作人員進行認證。我們致力於保護每一位創作者的權益。 點擊去認領

創作许可范围
在線生圖
進行融合
允许下载
商業许可范围
生成圖片可出售或用於商業目的
允许模型轉售或融合后出售
QR Code
下載SeaArt App
在移動端继續你的AI創作之旅