[ 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.44604232 [View]
File: 30 KB, 819x210, file.png [View same] [iqdb] [saucenao] [google]
44604232

>>44603058
>1. Use the NovelAI VAE
I switched over to that for these test images btw.
>2. Your negatives look like the AOM2-ish negatives.
I'm assuming the reason behind this is because it was merged with Anything v4 which actually is an AOM2 based model. Could this be contributing to the deep frying?
>Use whatever negatives the fat model recommends. If it doesn't recommend any, either find what negatives WD1.5 works best with, use the NAI ones HLL is a NAI retrain or mix them up. Newfriends don't remember what adding weight to negatives could do on NovelAI - deep fry shit.
I'll try and remove the weights from my negative prompts, but I think I have bigger problems considering Anything V4 is probably causing most of the deep fry.

>3. Turn your scale down.
I keep scale up high, and then rerun at lower cfg scale with highres for individual reruns.
>4. Try a different sampler? DPM++ SDE Karras 25 steps. Save you some time over Euler a
I actually didn't mean to use euler a intially, most of my deep fries have been happening with DPM++ 2S a Karras

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