@Dr. Furkan Gözükara i know you said that it wasnt practical, but is there any hope to see kosmos re
@Dr. Furkan Gözükara i know you said that it wasnt practical, but is there any hope to see kosmos replace llava in the supir app?
Athe SUPIR app was hastily thrown together by the developers of the upscaler and is pretty shoddy in general. Now that SUPIR is trending so much I’ll bet we start seeing more / different implementations, sort of like what Furkan has been up to
RI know it’s a long stretch of course. For now. But @Dr. Furkan Gözükara is up for feature requests this being the most popular project he’s put out yet u think, after his dreambooth
Ayeah, SUPIR is like going from Benadryl to Zyrtec as far as upscalers are concerned
Nguys
N
Ncan someone please explain why it does this to me
Ntrying to do a sdxl training
Nand just after it is catching latents.. it does this
Pdid you already configures correctly accelarate with accelerate config ?
Pyou must set mixed precision fp16 or bf16 , fp16 with you are running on kaggle
Pthen in kohya you need use same fp16 or b16 that you already defined in accelerate config
Ftesting newest SUPIR updates
Fno error
Fit uses defualt
Fthe SUPIR developers are being lazy. they duplicated UI into 3 UIs :/
F
Fnow i need to add this :d
F
Fnice working now time to add all new stuff
Phaha
CLooks neat
A@Dr. Furkan Gözükara Does your Patreon notebook still wok on Kaggle for kohya gui? https://www.youtube.com/watch?v=JF2P7BIUpIU&t=1251s
YouTubeSECourses
If you don't have a strong GPU for Stable Diffusion XL training then this is the tutorial you are looking for. We will use Kaggle free notebook to do Kohya SDXL LoRA training. In this tutorial you will master Kohya SDXL with Kaggle! 
Curious about training Kohya SDXL? Learn why Kaggle outshines Google Colab! We will uncover the power of free Ka...
Curious about training Kohya SDXL? Learn why Kaggle outshines Google Colab! We will uncover the power of free Ka...
Nbut it is not doing the training
Nlike.. it stops
Nsorry I just found the mistake. it was because i was out of cuda memory finally 
CHey everyone! Has anyone encountered an issue with Layer Diffusion where it doesn't render transparent material correctly when using LORAs? Here's an example: one without any LORAs, and one with a style LORA.
CWhat could be causing this and how can I fix this?
Fit work
Fi havent tested yet
F
Fadding face restore
Fright now
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dwhat version are we on>
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fi merged newest update
Fbut new features not yet
Fplease also leave a reply to here https://github.com/gradio-app/gradio/issues/7588
GitHub
I am implementing Gradio slider to my SUPIR app It works but in basic terms I checked documentation and I don't see how to make it full screen or even interactive = True not working It is looki...
D
Dfullscreen would be good but im happy you managed to implement it Doc
Fi am trying to add face improve to our gradiop
Dim confused Dr, if we dont use stage 1 why is it still an option, also the check box that will run Stage 1 before stage 2 again if we only use stage 1 what does this do?
Fthey added it for llava
Fbut i think useless
Di tried and failed to load sadtalker lol so I shall await your gradio.....(no pressure of course) 
Fsure
Dthe warning about the spatial transformer can be just ignored huh? when loading
Dv17 says v18 on gradio
Ddo i need to put anything in stage 1 output? if only using stage 2?