re update your kohya
re update your kohya
�How? I am using your installers for runpod
Freinstall
�Okay thanks
Fsingle layer quality is bad
�Okay I will test others as well. Is it possible to train multiple subjects in a checkpoint fine tune without bleeding?
Fstill bleeds
Fbut each concept can be trained on each different layer
Fprogressively
�Will try that
Fbut the quality is lower
Flol even single layer bleeds
Fdoesnt make sense
Dwhat do you mean by bleeding?
Fcharacters get mixed
For concepts
M@Dr. Furkan Gözükara yes its because flux is distilled
Myou cant fix the bleeding issue, even with lokr you still get bleed
Fi trained only single layer
F
Fsingle layer 10 specifically
Mflux is cooked, sd3 8b coming out soon hehe
Fi guess single layer training has no value except companies to give you cheap results
Msingle layer quality is bad
Mnot worth it
Fyep
Mi think there is an issue with using the new torch versions @Dr. Furkan Gözükara quality seems to change and training as well
Fwhich one?
Mthe new dev ones
Fi plan to test it actually
Fsince multi gpu for fine tune fails :/
Feven on 48 gb
Fcurrently i locked windows to 2.5 RC
Fi think it works fine
Mim not sure if its the new cudnns or the torch versions, but i noticed massive quality and learning drop off
Fcan totally happen
Mi am going to test as well rn 
M@Dr. Furkan Gözükara can having different cuda toolkit affect results?
Fi doubt that
Fi seen only different xformers in past
Fbroken training
Fso spda can do same
Myou're right i just tested and sdpa gives worse results
F48 gb multi gpu still fails :/
FFLUX: Gradient checkpointing enabled. CPU offload: True
number of trainable parameters: 11901408320
number of trainable parameters: 11901408320
Ftried latest accelerator and torch
Dwhy do you offload if you have 48 gb?
MPlease let me know if this isn't the right place for questions... I followed the tutorial for Swarm UI and to get Flux into Swarm UI. During the installation process I did have an error which resolved when I turned off my Nord VPN connection.
I am able to generate images just fine using Stable Diffusion 3, but when I try either flux model, I receive the following error:
Self-Start ComfyUI-0 on port 7821 failed. Restarting per configuration AutoRestart=true...
[BackendHandler] backend #0 failed to load model with error: System.Net.WebSockets.WebSocketException (0x80004005): The remote party closed the WebSocket connection without completing the close handshake.
Anyone know how to resolve? I've made sure both the Nord VPN app is disconnected and the Nord browser extension as well.
I am able to generate images just fine using Stable Diffusion 3, but when I try either flux model, I receive the following error:
Self-Start ComfyUI-0 on port 7821 failed. Restarting per configuration AutoRestart=true...
[BackendHandler] backend #0 failed to load model with error: System.Net.WebSockets.WebSocketException (0x80004005): The remote party closed the WebSocket connection without completing the close handshake.
Anyone know how to resolve? I've made sure both the Nord VPN app is disconnected and the Nord browser extension as well.
LHello, can you tell me how to reduce the scales of a trained lora flux? At my exit, lora is weighing 1gb
Z@Dr. Furkan Gözükara can models do like different font styles? or maybe do like font wildcards for making alogo?
Fmulti gpu doesnt work even with offload
