[ 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.38057487 [View]
File: 53 KB, 848x836, db.png [View same] [iqdb] [saucenao] [google]
38057487

>>38035443
Version posted was already merged with berrymix.

Also, if anyone wants to try dreambooth training, this settings work with 12gb VRAM. Dont forget to add COMMANDLINE_ARGS= --xformers to bat file. Recommended 20 images, 1800 steps takes 40 minutes on gtx3060. Not that bad

Good thing about dreambooth is that you can train on niche stuff that model doesn't understand by default. I've tried /h/-tier stuff like wooden_horse, chastity_belt and it works.
And with multiconcept and high gradient values it should be possible to make a proper /d/-tier/guro model. though using stuff like colossal ai is probably better for large dataset.

sorry for offtopic

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