does batch size really affects quality?
does batch size really affects quality?
AI think only 4640 and 3520, not 960.
KYeah I know. It's hard to read as hell. I'm using Chrome and still got this problem
KI'm not sure about other types of training, but in my style training (where I work with game characters in different forms like humans, humanoids, monsters, etc.), I noticed that increasing the batch size from 1 to 4 really makes a big difference in quality. So, I want to increase it even more. However, the RTX 3090 on RunPod can only handle a maximum batch size of 4.
KKinda liked 960 as well though
KIt had something
Fgo here
chrome://settings/fontsthat fixes it
Fbut improves it? Furkan just said that a bigger batch size reduces quality!
AYep, that was my understanding from his tutorials too.
AWhat is the recommended Python version for Dreambooth training again?
F3.10
AAny 3.10? or a specific point release?
Fany
AGreat, the RunPod template installs 3.10.10 but mine installed 3.10.6 so i was a bit concerned about that.
Fboth are ok
ASpot pods are saving me a lot of $ 
A
Fsoooo.. i've written a technical article about how stable diffusion works internally (but easy to understand for almost everyone)
Fthis is good to know because it helps understand better the most important parameters
FI appreciate if someone wants to read it and even more if give me feedback 
FUnderstand in a simple way how Stable Diffusion transforms a few words into a spectacular image.
Fyes
Fi think it reduces for teaching a person
Fthanks
Fi am currently focused on deep fake tutorial
F
Pi don't understand. Why are you showing us a blank video?
Fthis is face changed part. i will compare original and this in the youtube video
Fi was thinking this was the original lol
Fno you didnt watch superman :d
Fi didn't
Pit was a john cena joke. "you can't see him" -> blank video
Fhaha really :d
F
Fi am also showing face restoration
EHey! Quick question. Should we use Python 3.10.6 for stable diffusion or is 3.10.11 fine?
AAny 3.10.x version
Fpretty much this
Fcodeformer face fixing taking forever
F
Fi don't think it's a good thumbnail
Fit will be even smaller
Fit's already hard to spot the difference
Fmaybe zoom on the face?
Fhow about this one
F
Fexporting the video
Fbetter i think, but still.... faces are too similar xD
Fbut i approve! waiting for the video
waiting for the videoFwell these guys are similar :d
