Sign In

FromNewWorld (自新大陆)

78
421
7
Updated: Jan 14, 2024
base modelanime
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
421
Reviews
Uploaded
Jan 7, 2024
Base Model
SD 1.5
Usage Tips
Clip Skip: 2
Hash
AutoV2
C3970F872B
0
0
0
0

Introduction

本模型可以用于生成柔和明丽的色调和轮廓。

如以往出现的诸多优秀模型,该模型诞生的初衷是专门为自己定制的一种特别的画风,也可算作试验的产物,因为这是我第一次融合自己的底模型。但个人使用感觉良好,也十分欣赏这样的清淡风格,于是打算上传到本站。

这个模型不论是不使用任何LoRA还是使用一些角色LoRA,效果都是不错的。我的例图都是用的《明日方舟》和《原神》的角色。其他任意作品中的角色理论上都可以用该模型生成。

最后,你没猜错,不是说这名字有点令人耳熟,它就是取自德沃夏克的《自新大陆》。

This base model can generate gentle, beautiful and elegant colors and curves.

Like many other uploaders, I merged it initially for my personal use. But I appreciate it very much, so I would like to share it with others.

It's a good idea to use LoRAs of many characters when using this model. The example pictures I have uploaded are mainly related to Arknights and Genshin Impact. Theoretically, many LoRAs of other characters in various ACGN compositions can be adapted to this base model.

Finally, I like Dvorak, so I named it in this way.

How to Use the Model (just as a recommendation)

Basic settings:

VAE: kl-f8-anime2.ckpt/animevae

Steps: 20~30

Sampler: Euler a, Euler, DDIM

Resolution (Before HiRes): 512*768, 512*512, 768*512, 768*768

High resolution fix:

Hires upscaler: R-ESRGAN 4x+, R-ESRGAN 4x+ Anime6B (most frequently used), SwinIR_4x, ScuNET, ScuNET PSNR

Hires upscale: ≥2

Hires steps: ≥10

Denoising strength: around 0.25

CFG scale: 7~10

以上配置仅供参考,我猜甚至存在一些更加好的配置。但是话说回来,高清修复是一条命根子。

It is also okay to try this model with your own configurations. I do not think the quality will dissatisfy you a lot. But HIGH RESOLUTION FIX is vital, inevitable and indispensible.