Sign In

kimchiMix

738
6.5k
94
Verified:
SafeTensor
Type
Checkpoint Trained
Stats
1,672
Reviews
Published
Jun 4, 2024
Base Model
SD 1.5
Training
Steps: 20
Epochs: 100
Usage Tips
Clip Skip: 2
Hash
AutoV2
F470AC3947
default creator card background decoration
kimchi88's Avatar
kimchi88

CFG: 2.5! Too high CFG will cause explosion! [Hint word guide coefficient cfg scale]

Note: CFG is associated with reverse hint words (low quality:2) with equal weight. Increasing CFG requires appropriately reducing the weight of reverse hint words.

*1.5 The highest definition on the entire network, based on image algorithm-optimized direct output examples

When I output the images locally, I added some post-processing algorithms to make the images more interesting, which may result in some differences from other users when they are generating images.

"This algorithm mainly makes the image clearer, with a larger color gamut, and better details in the dark areas" — but these effects are unrelated to the model itself

Version: v4

Sampling method (Sampler): Euler Karras

Upscaling algorithm: 4x-UltraSharp

Sampling iteration steps (Steps): 20-30

HD restoration sampling times: 0-10

Redraw amplitude (Denoising): 0.2-0.4

Width 512, Height 768

Hint word relevance (CFG Scale): 2.5

======================================================

QQ group: 852450267

======================================================

建议CFG: 2.5!CFG高了会炸!【提示词引导系数 cfg scale

注意:CFG 关联反向提示词的(low quality:2)等权重。提高CFG需适当降低反向提示词权重

*1.5全网最高清,基于图像算法优化的直出样例

我在本地出图的时候加入了一些后处理算法让画面的表现更有意思,这会导致与其它用户在生图的时候有一些差异,

「这个算法主要使画面变得更清晰,色域更大,暗部细节更好」——但这些效果与模型本身无关

版本:v4

采样方法(Sampler): Euler Karras

放大算法:4x-UltraSharp

采样迭代步数(Steps):20-30

高清修复采样次数:0-10

重绘幅度(Denoising):0.2-0.4

宽度 512,高度 768

提示词相关性(CFG Scale): 2.5

======================================================

我的交流群:QQ 852450267

======================================================

版权说明:我的模型训练使用SD生图进行训练,所上传的模型成品在本人未允许的情况下不得用于数字模特、转载及搬运等商业用途。如果在商业用途中使用了该模型,那么一切风险与责任追究与本人无关,本人不承担任何法律责任!

Copyright note: My model training uses SD graphics for training, the uploaded finished model shall not be used for digital modeling, reprinting and handling without my permission. If the model is used in commercial use, then all risks and liability have nothing to do with me, I do not bear any legal responsibility!