Ai Hoshino
For V3 (soon)
For v2.5:
model: AbyssOrangeMixs 3, detailed project v5(on closertodeath profile on huggingface), animefull-latest, anything v4/3 Recommended VAEs: clearvae(best), klf8anime
tokens: ai hoshino
Description
More stable
& Better results
FAQ
Comments (23)
I'm a newbie here. I just copied arguments from this image but get overfitting image. Can anyone help me? thx
Model: dpepmkmp
Vae: ClearVae
Here is my screenshot and bad output
That's weird.. i haven't gotten this problem ever, could you try a normal generation without the same seed?
Ok i found your problem, it's because the upscaler
@KoRo_ Sorry for the long time no reply. I tested all type of latent upscaler, but always got overfitting image.
@kmou424 don't use latent, use R-ESRGAN or something else, latent ones doesn't work properly with it
@KoRo_ Thanks! I got a nice output🥰
@kmou424 i also found the actual problem
Latent upscaler requires a high denoising 0.6 and more
So it'll be noisy if you had lower than that, but well u can use other upscalers at the end anyway
for latent upscaling use a denoising strength of 0.54 or higher
@xyskl I like how exact this is 😂. You're not wrong though. Sometimes you can get away with 0.5 but that's pushing it. I always use 0.55.
非常抱歉...但是我不知道是什么问题
它总是看起来像这样吗？ 如果你不使用相同的种子？
别用additional network试试看，直接<lora:aiHoshinoLycorisOshiNoKo_v25:0.7>。
我以前对比过，出来的图不一样。
@KoRo_ DPM++ 2M Karras v2大佬你好请问这个Sampler是在哪里下的
@weimo1997 我不知道。我用过 NoCrypt Google Colab 去测试
@weimo1997 https://colab.research.google.com/drive/1wEa-tS10h4LlDykd87TF5zzpXIIQoCmq?authuser=9#scrollTo=8Uvqba0yKW7U
Can I know where to download Lora:kyotoanimation_v10 please.
Thanks!!
复活吧！我的爱人！