Software Engineering Courses (SECourses)•16mo ago

You can use the layers he mentions here - https://www.reddit.com/r/StableDiffusion/comments/1f523bd/

You can use the layers he mentions here - https://www.reddit.com/r/StableDiffusion/comments/1f523bd/good_flux_loras_can_be_less_than_45mb_128_dim/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

From the StableDiffusion community on Reddit: Good Flux LoRAs can b...

Explore this post and more from the StableDiffusion community

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 You can use the layers he mentions here - https://www.reddit.com/r/StableDiffusi...

Furkan Gözükara SECourses•9/17/24, 11:14 AM

i tested 0 to 7 none looks a like

Furkan Gözükara SECourses•9/17/24, 11:14 AM

i think new LR is mandatory

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮OP•9/17/24, 11:19 AM

might be the case then

Tacite•9/17/24, 11:52 AM

training on layer 7 and layer 20 is not too bad. Using noise_scheduler: flowmatch, optimizer: adamw8bit and lr: 0.0005

TTacite training on layer 7 and layer 20 is not too bad. Using noise_scheduler: flowmat...

Furkan Gözükara SECourses•9/17/24, 12:59 PM

ye your LR is 10x of mine

Furkan Gözükara SECourses•9/17/24, 12:59 PM

i am doing more trainings

Furkan Gözükara SECourses•9/17/24, 12:59 PM

atm

Zono50•9/17/24, 1:18 PM

@Dr. Furkan Gözükara so supir tool was built using cudatoolkit 11.8? but xformers is looking for pytorch 2.4.0 +cu121, i'm confused

Zono50•9/17/24, 1:19 PM

not sure why it says that cause i don't even have pytorch 2.2.0 installed i have 2.4.0

PixelMonkey•9/17/24, 1:20 PM

Question in the round of you awesome nerds

. Iw as trying the last days control net with flux in SD Forge webUi but it seems broken.. neither the Depoth or the Canny mode worked for me .. Any one having the same issue ?

ZZono50 @Dr. Furkan Gözükara so supir tool was built using cudatoolkit 11.8? but xforme...

Furkan Gözükara SECourses•9/17/24, 1:24 PM

use our installer it handles all

FFurkan Gözükara SECourses Click to see attachment

smoke•9/17/24, 1:33 PM

Will you be doing finetune vs lora comparisons?

Ssmoke Will you be doing finetune vs lora comparisons?

Furkan Gözükara SECourses•9/17/24, 1:35 PM

yep i already did too

smoke•9/17/24, 1:36 PM

Oh you already did?

Furkan Gözükara SECourses•9/17/24, 1:36 PM

check images

Furkan Gözükara SECourses•9/17/24, 1:36 PM

https://www.patreon.com/posts/full-fine-tuning-112157789

Patreon

Full Fine Tuning / DreamBooth of FLUX yields way better results tha...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

Furkan Gözükara SECourses•9/17/24, 1:36 PM

left best lora right fine tune

smoke•9/17/24, 1:37 PM

Hmmm I see, finetune better for sure. Have you also tried finetune lora extract? I tried that but the quality dropped drastically, something was definitely wrong with the lora extract script

Ssmoke Hmmm I see, finetune better for sure. Have you also tried finetune lora extract?...

Furkan Gözükara SECourses•9/17/24, 1:59 PM

not tried yet

Furkan Gözükara SECourses•9/17/24, 1:59 PM

but it should be better than lora training it self

smoke•9/17/24, 2:00 PM

Yeah I also think that. I'd use it but that lora extract script is fully broken, atleast for me. And keeping those huge checkpoints is a bit much

FFurkan Gözükara SECourses ye your LR is 10x of mine 😄

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮OP•9/17/24, 2:12 PM

what network dim and alpha are you using?

Kallamamran•9/17/24, 3:05 PM

@Dr. Furkan Gözükara I can't seem to find that captioning editor GUI you've made. Would you mind linking to it?

FFurkan Gözükara SECourses i tested 0 to 7 none looks a like

dsienra•9/17/24, 3:34 PM

I want to try this but I have some questions, layes=bloks what's the diference between single and double blocks? you are testing with single blocks or layers? can I test with 7 and 20 using a higher lr, let us know your findings, thanks

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 what network dim and alpha are you using?

Furkan Gözükara SECourses•9/17/24, 3:56 PM

128x128

KKallamamran @Dr. Furkan Gözükara I can't seem to find that captioning editor GUI you've made...

Furkan Gözükara SECourses•9/17/24, 3:58 PM

https://www.patreon.com/posts/108992085

Patreon

Image Captioning Editor Gradio APP - Edit Your Captions Super Easy ...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

Ddsienra I want to try this but I have some questions, layes=bloks what's the diference...

Furkan Gözükara SECourses•9/17/24, 3:58 PM

that is what i am testing now :d

Furkan Gözükara SECourses•9/17/24, 3:58 PM

done 16 trainings so far training 8 more

FFurkan Gözükara SECourses 128x128

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮OP•9/17/24, 4:07 PM

You can try 256 if you don't see good results

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 You can try 256 if you don't see good results

Furkan Gözükara SECourses•9/17/24, 4:07 PM

yes can be improved nice idea to test

Darth Teethius•9/17/24, 4:55 PM

looks like individual block training might have been implemented in the sd3 flux1 kohya ss branch last night

xlads•9/17/24, 4:55 PM

So fine tuning is better the Lora over all you think @Dr. Furkan Gözükara ?

Darth Teethius•9/17/24, 4:56 PM

DDarth Teethius looks like individual block training might have been implemented in the sd3 flux...

Furkan Gözükara SECourses•9/17/24, 4:56 PM

i am testing already

Xxlads So fine tuning is better the Lora over all you think @Dr. Furkan Gözükara ?

Furkan Gözükara SECourses•9/17/24, 4:56 PM

yep

xlads•9/17/24, 4:57 PM

You have a video on how to fine tune training?

Ra25•9/17/24, 5:22 PM

Can someone confirm Flux Lora Extraction in completely broken on Kohya ?

Ra25•9/17/24, 5:25 PM

I did a finetune of flux dev, results are amazing. I use Extract Flux LoRA tab to extract the Lora from the finetuned model. When I use the extracted Lora it makes no difference with not using Lora. @Dr. Furkan Gözükara Did you try the Extract Flux LoRA feature ?

xlads•9/17/24, 5:40 PM

I’m confused on what fine tuning is, do we basically do the same training as in this video https://youtu.be/nySGu12Y05k?feature=shared but with the configuration from the new update on your patreon?

YouTubeSECourses

FLUX LoRA Training Simplified: From Zero to Hero with Kohya SS GUI ...

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...

RRa25 Can someone confirm Flux Lora Extraction in completely broken on Kohya ?

Neo•9/17/24, 6:10 PM

I'd also like to know. Though I was waiting on Kohya to refine it a bit more before delving into fine tunes.

RRa25 Can someone confirm Flux Lora Extraction in completely broken on Kohya ?

smoke•9/17/24, 6:11 PM

Same here. Do you mean extracting a lora from a finetuned checkpoint?

Neo•9/17/24, 6:12 PM

Yeah

smoke•9/17/24, 6:12 PM

For me the extracted lora looks nothing like my subject, nothing like the checkpoint

smoke•9/17/24, 6:12 PM

I guess it is broken

Ra25•9/17/24, 7:26 PM

@smoke yeah, thx for the confirmation

smoke•9/17/24, 7:28 PM

Should probably raise an issue in sd-scripts Github since it is broken

Ra25•9/17/24, 7:35 PM

There's already one https://github.com/kohya-ss/sd-scripts/issues/1496

GitHub

Extracted flux lora not working · Issue #1496 · kohya-ss/sd-scripts

2 Issues sometimes it works, but it looks nothing like the fine tuned model. Doesnt work and i get the error below During handling of the above exception, another exception occurred: Traceback (mos...

smoke•9/17/24, 7:36 PM

Ah yeah it definitely is broken

Ra25•9/17/24, 7:36 PM

But i'm not sure it's the same issue

Ra25•9/17/24, 7:37 PM

Because they say it works at 1.5 strength

You can use the layers he mentions here - https://www.reddit.com/r/StableDiffusion/comments/1f523bd/

Similar Threads

Similar Threads

Similar Threads