[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/vt/ - Virtual Youtubers

Search:


View post   

>> No.47102421 [View]
File: 1.77 MB, 968x1336, 00018.png [View same] [iqdb] [saucenao] [google]
47102421

>>47101842
I don't know what I'd suggest for learning rates, it's all trial and error anyway. Also, if you have a trigger word for the lora you would want your keep tokens to be 1 and for shuffle caption to be on. I hear that people also use flip augmentation for low datasets but results may vary. Also, if you could give some tags from the dataset i think I could probably test the lora a bit more.

Navigation
View posts[+24][+48][+96]