So to confirm, with a 5090 you recommend keeping batch size 1 and acc. steps 1 as well correct? I di
So to confirm, with a 5090 you recommend keeping batch size 1 and acc. steps 1 as well correct? I didn't know it reduced quality
Fall 1
Falso use newest tier
Ftier1_17_GB_fast_v4.json
KWill give it a shot and let you know if perhaps that resolves it! Thanks
�on windows when i run kohya installer, i hit option 1 to install, once i press ENTER, the terminal just closes
�installation was complete in like 15 seconds, obviously it did not install
Fhow to change lr_shceduler to cosine in fluxgym? its default to constant with warmup
Fyou are lacking python
FYouTubeSECourses
If you are interested in using AI, generative AI applications, open source applications in your computer, than this is the most fundamental and important tutorial that you need. In this tutorial I show and explain how to properly install appropriate Python versions accurately, how to switch between different Python versions, how to install diffe...
Fi never used fluxgym
Fwhat do you suggest using?
Ffor lora training
Five 48gb vram but with fluxgym im unable to fully utilize it. its only using 20gb vram
Fkohya
Ffor 48 gb we have batch size 7 dreambooth config
Ffully utilizes the GPU
FYouTubeSECourses
If you want to train FLUX with maximum possible quality, this is the tutorial looking for. In this comprehensive tutorial, you will learn how to install Kohya GUI and use it to fully Fine-Tune / DreamBooth FLUX model. After that how to use SwarmUI to compare generated checkpoints / models and find the very best one to generate most amazing image...
Fif you want lora it is here https://youtu.be/nySGu12Y05k
YouTubeSECourses
Ultimate Kohya GUI FLUX LoRA training tutorial. This tutorial is product of non-stop 9 days research and training. I have trained over 73 FLUX LoRA models and analyzed all to prepare this tutorial video. The research still going on and hopefully the results will be significantly improved and latest configs and findings will be shared. Please wat...
GHave an 3090 and recently did a fresh install of Kohya, followed by 'Windows_RTX5000_Series_Upgrade_Run_After_Install_Finished.bat', and previously my speed was ~8s/it but now its ~11s/it. Tried reverting and doing fresh installs again but the same issue. Any idea on what's the best way to debug what's causing this?
Fis your dataset all 1024x1024?
Gmy actual images are larger, but my 'Max resolution' is set to 1024x1024 (so assuming they're getting resized to 1024x1024 themselves)
Fthat is why
Fit wont resize to 1024x1024
Fit will only resize small side to 1024
Gsorry, I should have said that they're all 1:1 aspect ratio, so would end up being 1024x1024
Fhow much watt your gpu using did you check?
Fi mean like this
F
Gjust rerunning it now - 2 mins 
G^not sure if that is comfyui specific and not related to Kohya though
G
Fyou need to increase block swap count
Gah ok, I haven't changed the config at all from when it was running at 8s/it. I've already tried blocks to swap ranging from 12 - 22 in each of the 3 set-ups below, but doesn't seem to get it past 11s/it:
- Old install + old config + 'Windows_RTX5000_Series_Upgrade_Run_After_Install_Finished.bat' file -> 11s/it
- New install + old config -> 11s/it
- New install + new config -> 11s/it
- New install + old config + 'Windows_RTX5000_Series_Upgrade_Run_After_Install_Finished.bat' -> 11s/it
- New install + new config + 'Windows_RTX5000_Series_Upgrade_Run_After_Install_Finished.bat' -> 11s/it
Fare you waiting at least 100 steps
Fto see real speed?
Gyup, i'm waiting for it to bottom out. Over 3000 steps from run from last night:
Fso even you change block swap
Fyou cant get better?
Fyour watt usage is low
F
Flet me test mine
Fthis is dreambooth or lora?
Gdreambooth
Gyup, that's right
CHey @Furkan Gözükara SECourses I hope you're doing great, just wanted to thank you for your latest tutorials and also ask if the video upscaler is still in the works coming soon, or you are now prioritizing other stuff (really hard to catch up with ai nowadays). After I saw your latest video upscaler new implementation posts I'm really looking forward to test it out.
SFor flux lora training. I have images 2046 * 3050px size but it needs 1024 * 1024px. What should I do?
GAny idea on this Furkan? I've just done another fresh install + latest cudnn + 'Windows_RTX5000_Series_Upgrade_Run_After_Install_Finished.bat' + Cuda 12.9. Running on a fresh windows reboot (so no background process), block swap set to 22, and still limited to 11.37s/it
really confused on what else I can try. Just odd how I was getting 8s/it beforehand and now with more up-to-date packages its running slower. I can't really revert it either, so would love to know what you'd try to do next in my situation
really confused on what else I can try. Just odd how I was getting 8s/it beforehand and now with more up-to-date packages its running slower. I can't really revert it either, so would love to know what you'd try to do next in my situation
Fye here better to write
