Suggestion/建议:
Step 1/第1步: txt2img/文生图:
Negative prompt: (worst quality:2), (low quality:2), (normal quality:2)
Sampling steps/迭代步数: 40-50
Hires. fix/高分辨率修复: On/开
Upscaler/放大算法: R-ESRGAN 4x+ or 4x-UltraSharp
Hires steps/高分迭代步数: 20
Denoising strength/重绘幅度: 0.2-0.4
Step 2/第2步: Send to img2img/>> 图生图:
Resize mode/缩放模式: Just resize/仅调整大小
Denoising strength/重绘幅度: 0.2-0.25
Script/脚本: SD upscale/使用SD放大(Upscaler/放大算法:R-ESRGAN 4x+ or 4x-UltraSharp)
请勿商用
Description
Full version based on SP-Light.
FAQ
Comments (6)
TMND-Mix整个系列我都超级喜欢
Hey,
I love this model and just upgraded to the VIIBakedvea version. I have a question someone can maybe help with.
The outputs are stunning and high quality, but are very biased towards a soft pink/purple/red color palette (see my imagine post for images with generation data).
Does anyone know what is causing this? It is not present in the sample pictures and other posts for this model.
@gxf159357233 Thank you for your response. The VAE you suggested does give great outputs, but sadly doesn't remove the bias towards the red colors.
I found that increasing the clip skip to 3-4 or higher and prompting for colors like blue and green does help, but also does not quite give natural colors for me.
@Bloo That's probably because of the style of the model, which produces well-lit scene images colors that tend to be red. Maybe you can use some saturation lora or color lora to improve it.
Like this https://civitai.com/models/102639?modelVersionId=109827
@gxf159357233 Thank you for responding again. These color balance Loras work quite well with helping to balance the red levels of my outputs, with minimal effect on the image/style.
Thank you for showing these to me, I'll use them in the future. Keep up the great work, the SP Rainbow Version is also looking fantastic.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.