Thanks! Is there any whitepaper available about this theme?
Thanks! Is there any whitepaper available about this theme?
Fnot that i know
Fbut flux works best with 1024x1024 i find that
Fbest is comparing
F1024 vs bucketing
V@Furkan Gözükara SECourses I want to train a flux lora based on oil paintings. I just want the lora to capture the textures, brushstrokes and art style, not necessarily the colors. Would fully captioning the dataset be preferred over using an activation word?
T@Furkan Gözükara SECourses I'm a bit confused, the youtube video says to load the configs into the lora tab, but with the latest best configv5 they load into the dreambooth tab. Which tab do you do the config in, and which tab do you hit start training? (UC Arcade on patreon)
SSame thing is happening on my GTX4090. You in Windows 10? I'm about to nuke the computer and throw Ubuntu on it because I think Windows is the problem
VCheck to see if its using shared memory, if so that might be the issue, which means the config needs tweaking for your machine.
SYeah, I've tried everything everything google will tell me to kill the shared memory. Gonna try the 16gb config file and see how that goes before I blow up the OS
Fi think you can still train without captions
Fand define the colors during inference
Fnew configs are dreambooth
Fa video is not made yet
Fbut exactly same
ABucketing means that the system caches them internally to the best fit?
ALike when using AI-Toolkit?
A
X@Furkan Gözükara SECourses what is happening if we train transparent images? like lets say we got icons but background removed png, will it be alright?
Fyes
Fit will put them into buckets
Fand auto crop and resize
Fit will be black background
Fduring training as far as i know
For maybe white 
Ayep, depends on your settings, I write a little padding program atm, utilizing the graphics32 library for batch processing images to maintain the 1024x1024 ratio.
WIn terms of Kohya FLUX Fine Tuning, is it possible to do it on any Flux D models? Or must we use the original regular Flux Dev checkpoint? @Furkan Gözükara SECourses
U
Fuse original. but i will research distilled ones too
Fdidnt test yet
WI appreciate the response. thank you.
I trained a flux Lora on an anime versions of my original character.
The likeness and fidelity is excellent.
But it keeps trying to create 3D and semi-realistic versions of my character.
It's a very difficult to get it to output in flat shaded anime style.
So I am hoping that dream booth training can fix that problem.
I trained a flux Lora on an anime versions of my original character.
The likeness and fidelity is excellent.
But it keeps trying to create 3D and semi-realistic versions of my character.
It's a very difficult to get it to output in flat shaded anime style.
So I am hoping that dream booth training can fix that problem.
Fit would help 100%
Fso whatever youy get with lora
Fyou will get better with fine tuning
WDoes your recent patreon post contain instructions to do it on runpod? If it does, I will rejoin as a member for a couple months
Fyep. only difference is
Fload config into dreambooth tab
Frest 100% same
Feach checkpoint will be 23.8 gb pay attention to storage
Fi am also preparing a massive tutorial
W23.8gb. Holy crap!
Is it possible to extract a lora from that? is it worth it?
Is it possible to extract a lora from that? is it worth it?
Fit works
Fbut you still need high dimension :d
Fi made article here public
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Wthankyou. I will wait for the runpod tutorial.
FYouTubeSECourses
Unlock the power of FLUX LoRA training, even if you're short on GPUs or looking to boost speed and scale! This comprehensive guide takes you from novice to expert, showing you how to use Kohya GUI for creating top-notch FLUX LoRAs in the cloud. We'll cover everything: maximizing quality, optimizing speed, and finding the best deals. With our exc...
Fstart from 44:38
Fi wont repeat all again
Fin new tutorial cuz exactly same
Wso this includes the dreambooth tutorial?