will smith eating a spagetti western
will smith eating a spagetti western
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Cwill smith and furkan playing chess
Fnot working atm. but i am doing new onetrainer training
Flets see if will fix
Cyou can't prompt to have will smith join you at a chess table?
Fnope not working
Fit bleeds
Cnot surprizing since you're using flux. try it with sdxl
Fyes flux bleeds
Fsdxl should work
Cthat's because it's not trainable. let's see what SDXL does with you and will smith
XThis is cool!
XCan you make a tutorial for how to fine-tune?
OJust a quick question on Supir- it's amazing. Running locally on a RTX3080, and it takes forever. Can you use multi-gpu with Supir, so I can throw batches up on Massed compute? Is the speed more a function of gpu compute or vram?
EI'm quite happy with it, not to mention it looks fancy, got a G.Skill Trident Z Royal kit cuz I figured if I'm gonna put out the money for 96GB of RAM might as well get the fancy looking one with the good timings lol
Question though, how much RAM usage do you standardly see while running Flux?
Question though, how much RAM usage do you standardly see while running Flux?
EIt is bright as hell though, and needs active cooling for what settings in pushing it to already. Haven't stability tested it yet but I'll do that on my days off. So far it seems rock sold. I'm only doing CL30 at 6200MHz 1.35v, the kit is rated for CL32 at 6400MHz 1.35v but I've yet to get 6400 to run stable because of AMD's fun memory controller stuff.
Done problem I'm having with supir is that some times makes incorrect teeth
Dworks quite good on 3090
DI don't know about multi gpu
Dcan I tell Supir to load a checkpoint from a folder outside the supir installation?
Dnvm found it
Dis there a specific reason why EDM is the default sampler for SUPIR?
DPMPP2M looks better to me, especially for skin. render time is the same too
https://imgsli.com/MzAzMTk3/
https://imgsli.com/MzAzMTk5
DPMPP2M looks better to me, especially for skin. render time is the same too
https://imgsli.com/MzAzMTk3/
https://imgsli.com/MzAzMTk5
AI am using this tutorial in runpod, when i try to load "Rank_1_28700MB_Slow.json" i get connection error and i can't proceed any further. No matter how many times i try.
APatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fyep i am trying to perfect it
Fon massed compute
Fskip tiling
Fit hugely speeds up
Fram usage shouldnt be much unless your vram is low
Fbecause when vram is low cpu off loading is made
Fand it uses a lot of ram
Fwell EDM is more loyal but you can use both 
Factually we are working on adding a lot of new stuff too
Fhi are you using gradio live share?
Fvery likely your runpod pod is broken
Fi would prefer massed compute over runpod anytime
AApreciate your prompt response, i am using direct http link from pod
AI will give Massed compute a try.
Fah yes that is why failing
Frunpod proxy
Fwow this is really great for flux
F
BAnyone here got JoyCaption working in ComfyUI?
OJust to clarify: no multi-gpu, but an A6000 can skip tiling because it has 48gb vram?
OIs this the result of the 256 image data set + 7 batch size?
F
Fit is around 1460 steps and 1 step is around 30 seconds
Fon single A6000
Fso cost is around 3 4 usd