Regarding the noise setting

[tellurion-kanata]

Hi, I would like to know how you determined the linear start and end points, which are typically set as 0.00085 and 0.0120. What are the advantages of using these values compared to others, such as the ones used in the semantic synthesis sd-v1.0 model, which starts at 0.0015 and ends at 0.0205?

I am using this setting to fine-tune the model on my own dataset, but I have noticed that the generated images are usually darker in hue. Increasing the noise level may improve this issue, but I am concerned that it may cause other problems or degrade the final performance.

Thank you!

[ming1993li]

Hi, do you have the answers?