is 3x or 4x possible with 24 gb vram? (from 1344x768)
is 3x or 4x possible with 24 gb vram? (from 1344x768)
Kmax i go is 2.5x
Fi am going to add it soon hopefully working
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dyeh x3 and x4 might be pushing it
Kthank you 
Dat that resolution
Rcant say there is a noticeable difference between stage1+2 and stage2
Rit might be a subjective call betwen 1+2 and 2 only
Ri am going to stick to stage2 only
Fye i will make it optional
DI was reading somewhere that the prompting makes a big difference and that it can change materials by prompting suede and adding things like a moustache, can turn a person young / old male / female
Fi will add stage 1 + stage 2 button as well
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fyes it has impact
Fgiving stage 1 input in stage 2
Fyielded very bad image
Fare you testing that way?
P4x OOMed but 3x using 21/24 gb vram rendered in 246.27 seconds (power limit is set to 60%)
Fwhat is final res
F
Fv17 is live
F
Rcool. what is the number of images?
Flets say if you make 3
Fit will generate 3 images with 3 seeds
F1 by 1
Roh i see
Rbut for upscaling, is that that useful? in magnific i can see the use, because it is meant to hallucinate and be creative with new details
Rbut here?
Ris it for testing purposes? to see what seed gives you best upscale?
Fsometimes some seeds
Fyields better
Fthan others
Fyou know like in automatic1111
Fbut not that dramatic
Ri see
Rthat would be crazy to start upscaling multiple seeds. it's already time costly
Rdo you forsee faster quantization coming soon?
Fauthors are super active
Fi expect improvements
Fthey are adding face restore
Ri noticed an a5000 is 0.21 $ for 20gb vram compared to a6000 0.79$ 48gb. Its 1/3 price but about only 30% slower (with optimisation)
Rso not sure if good option, for price
R3090 is faster than a6000 older generation, but with limited vram and using optimisation it is a lot slower than a6000
CI do multiple seeds for really terrible pictures. Like I got pics taken with mid 2000's phones that are just a blurry mess and pretty much broken beyond repair. But I feel like SUPIR might be the way. I did some experiment with SUPIR then Reactor... The results arent natural, but it feels like a legit decent start.
Fye the prices not reflect the speed
R@Dr. Furkan Gözükara can you possibly add more information in the console during process? On runpod i am only seeing this
Ris there a way to run a tail to see more of the process?
Frequires to modify their pipeline which i dont want to
Fbecause it would make it harder to merge once they improve