It seems they were purposely censored just as some other anatomy parts during base model training
It seems they were purposely censored just as some other anatomy parts during base model training
JGot it
Jsome other parts
XI'm starting to think that it might be smarter for me to do wan inside of massed compute because even if i have to download wan every time i will use it inside of massedcompute it will still be fast and cheaper the runpod 
Jprices are almost the same?
Xmaybe but runpod has so many issue
Xfor now i will stick with runpod since i already have money in there but i think soon i will try to move to massedcommput
JBeen using Runpod a lot of past few days, works (almost) as well as Massedcompute IMO, but the templates just makes everything 20x easier to setup
Jhmm yea dont know about that
Jbut from my testing the RTX 4000 Ada and 2000 are kinda shit
JA40 is 2x faster on Runpod from my tests
Ni'm seeing about a 25% increase in speed in the latest update to the WAN app on my 3090 
XHmm i think I might have to try it out
X@Dr. Furkan Gözükara my 720p image to video took around 2 hour to genrate on this kind of pod
XCan you recommend a pod that can help me generate wan much faster?
FH100
Frtx 4090 would be faster too
For L40S
SDr Furkan how many as increase the speed with teacache?
Ximage to video 720
Fit depends on your threshold
Fbigger faster
Fbut lower quality
Feven better
Jteacache update is not great here. I hope it will be faster in the end
Non my 3090 i was getting about 60s/it in the early steps but starting at about step 35 it started getting faster and was down to 33s/it for the last 4 or 5 steps
Jnot getting better. I'm using img to video 480p with 16gb VRAM and 32gb RAM (+100 GB of virtual RAM)
JI think it's even worse than without teacache
Fit gets faster as progress
Fwait
Fi am coding multiple lora and new auto scale feature now
Fby the way i hope original authors adds better teacache than me 
Jnot getting faster
Fit gets eventually you will see
Fif you want even faster you have to use higher
Fbut it reduces quality a lot
Flike 0.2 increases speed over 100%
JI will try FP8 later to see if with v30 is fixed for me
JIt's getting worse
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Did go for the 480 model and upscale the vid using other software, 740 is so slowwwww
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fhopefully upscale coming :d
F
Jit's true
SDr Furkan what do you think could be a solid value treshold?
XIs there version of wan online? That people can just pay for that’s as good as the one I’m running in runpod?
DIm waiting for Wan, 2, 3
Dlil humour, yw