[ 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.49734068 [View]
File: 32 KB, 477x467, file.png [View same] [iqdb] [saucenao] [google]
49734068

>>49733876
CENTER OPENING MY BELOVED
I take it the interdimensional skirt lift was also unprompted/a jacket negative side effect?
>>49733869
hmm, maybe a similar argument for/against schizoprompting then? at least in my experience, especially when I'm asking it for help with code, I have to go out of my way to tell it what I do NOT want up front - because trying to tard wrangle it into doing that after the fact will make it "forget" previous stuff if it takes too many add'l messages to get my point across it told me flat out that it couldn't remember my storage topology when I was asking it for help with ZFS/RAID/Windows Storage Spaces. ofc it might've been hallucinating about that too, but it said it has a "sliding window" of conversation history. it being ChatGPT3, not Bing in this case
>>49733914
>what values do you even use for SAG and dynamic thresholding?
I YOLO them, kek. I've only used SAG a few times so I have it at default, which is I think 0.75 & 1. I've tried it at 1 & 1 once as well. DT is a bit more complicated, but I always set the mimic CFG scale to half of whatever I have scale actually set to. and clamp latents gets set to a random radio station from my hometown, lmfao
increase the clamp latent percent as you increase your actual CFG. you shouldn't really be going much higher than ~95-97% if you're not past cfg actually being set to 30. I'll adjust those 4's up a bit if it's a higher scale value, like 4.5 or 5.
this combination of schedulers and values keeps it from deep frying on SDE

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