Fmaking runpod massed compute and kaggle to publishj on patreon
Fdmn A40 is cheap on runpod for some reason
F48 gb
APineapple cat lol
Adr have you trained loras for Flux yet
Fnot yet sadly
Fi am waiting kohya to finalize
Asimpletrainer I think some people are using
Ahave you tried it
Asimple tuner*
Fyes ,i follow it
Fbut i am waiting kohya
Fit is gonna use hugging face version
Flol look at vram
F
Flol filled entire shared vram haha 
Fbut works great
Fas much as you can fit
Ffor high res
Fi am waiting next gen gpus to move to 128 gb faster RAM
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dwhat is the difference between comfui installer v3 and v4?
Dapart from the obvious 
AWhats hugging face version mean?
Fi mean diffusers : https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_flux.md
GitHub
Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX. - huggingface/diffusers
Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX. - huggingface/diffusers![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fi dont remember check info
Dlol
Ais there a downside to simple tuner?
Fharder to use. i dont know if working accurate never tested.
VShouldn't that be expected?
VWhen even 24gb dedicated ram becomes less
.
.VDoc we need to do a GPU heist at the Nvidia headquarters .
.
Fwe need
NThank you so much for this, you've optimised one of my processes around image processing so much with this!!!!
NI don't think 5090 will have that much VRAM, 48GB at max, I dont even think nvidia will go that route. Truly cloud computing for this stuff is the future-- and I am saying this as a 3090 and 4090 owner 
F48 gb would be amazing
Frumos are like 28 32 which are totally lame shameless
Fyou are welcome. this is the best atm
NThese examples are extremely good (at work so can't test). Was this using the general model or the human specialised model? https://huggingface.co/datasets/schirrmacher/humans
BI'm making a lot of grids for Very Special characters, so I built a quick web app to speed up evaluating them! I hope you find this free tool useful. https://everlyheights.tv/tools/stable-diffusion-tools/everly-heights-xyz-grid-evaluator/
Everly Heights | Stories set in the fictional town of Everly Heights, OhioEverly Heights
The Everly Heights XYZ Grid Evaluator is a powerful web-based tool designed to help you evaluate and rank outputs from your Stable Diffusion AI image models after you generate an XYZ Plot/Grid to test various settings, including LoRAs (Low-Rank Adaptation) and more technical settings like Steps and CFG. Whether you're an AI researcher, developer...
KNeeded in ComfyUI
KAll these stand-alone "apps" are cool and all, but it's not really useful jumping between all the different apps Upscaling is fine though sice that's usually a last stage
Upscaling is fine though sice that's usually a last stage
Fhi all, do we have any comfyui workflow to train flux on a character (like using dreambooth)
KNo. Not yet. Demands to much VRAM so noone has put any effort into that
KFlux trainings is done on rental-GPU's for now
Fokay i see, too bad!
FPS: you can run any comfyui workflow on rented GPUs on replicate: https://replicate.com/fofr/any-comfyui-workflow
Mhey doc, any toughts on how can i make the supir and vton app serverless after i deployed into runpod ?
Hinteresting.. can you run lora models as well?
FIs there a way to save your state from Massed Compute? It's kind of annoying that I have to reinstall all the custom nodes and models I wanna use everytime I boot an instance. Runpod's nice that I can keep a storage instance for cheap/hr
IHi guys, there's any tuto for loras on flux?