Software Engineering Courses (SECourses)•2y ago

Yes, it did not work with both v8 and v9, let me know if there’s a working version. I even tried Rea

Yes, it did not work with both v8 and v9, let me know if there’s a working version. I even tried RealvisXL_v4 too and got the same error: AttributeError: 'NoneType' object has no attribute 'tokenize’

FFurkan Gözükara SECourses bleeding happens for 2 reasons

dsienra•7/29/24, 9:57 AM

on xl I trained many loras and finetunes many subjects of the same class and no bleeding, the model learns the diference between person1 man and person 2 man, in my sd3 tests all bleeds, is like is no learning that the combination of person1 man is diferent of person2 man, it seems to learning all mixed no mather what token or class Im using

dsienra•7/29/24, 10:01 AM

Lets see with TE enabled, I will report in some hours when the new training is finished

MMarc-Daniel Schulz senor Gözükara, I sag your post on CogVLM 2, and im now wondering if about capti...

Furkan Gözükara SECourses•7/29/24, 10:07 AM

well best would be how users gonna prompt

Ddsienra with your setting for xl I can train many subjets of the same class with no blee...

Furkan Gözükara SECourses•7/29/24, 10:08 AM

wow nice

___pirate_king__Yes, it did not work with both v8 and v9, let me know if there’s a working versi...

Furkan Gözükara SECourses•7/29/24, 10:08 AM

did you enable SDXL checkbox?

Ddsienra on xl I trained many loras and finetunes many subjects of the same class and no ...

Furkan Gözükara SECourses•7/29/24, 10:08 AM

it could be also due to your training config

Furkan Gözükara SECourses•7/29/24, 10:08 AM

the config makes difference

FFurkan Gözükara SECourses did you enable SDXL checkbox?

__pirate_king__OP•7/29/24, 10:10 AM

Yes, also how would I know what ip-adapter it’s using. And is there any way we could upload multiple images as input from a folder?

___pirate_king__Yes, also how would I know what ip-adapter it’s using. And is there any way we c...

Furkan Gözükara SECourses•7/29/24, 10:13 AM

multiple images are getting averaged but currently we dont have it

Furkan Gözükara SECourses•7/29/24, 10:13 AM

what you are trying to do exactly?

Furkan Gözükara SECourses•7/29/24, 10:14 AM

when i have time i will test if it actually fails to convert diffusers

__pirate_king__OP•7/29/24, 10:15 AM

I tried doing it by converting the safetensors which didn’t work, now I’ve tried by adding the model id from hugging face into the list and it works for realvisXL.
I need to upload multiple images and create a virtual tryon image generation

___pirate_king__I tried doing it by converting the safetensors which didn’t work, now I’ve tried...

Furkan Gözükara SECourses•7/29/24, 10:22 AM

for virtual try on

Furkan Gözükara SECourses•7/29/24, 10:22 AM

IDM vton is best

FFurkan Gözükara SECourses IDM vton is best

__pirate_king__OP•7/29/24, 10:26 AM

Can we integrate that into this ?

___pirate_king__Can we integrate that into this ?

Furkan Gözükara SECourses•7/29/24, 10:29 AM

you can if you know programming :d

Furkan Gözükara SECourses•7/29/24, 10:29 AM

but already have idm vton app why dont you check it?

Furkan Gözükara SECourses•7/29/24, 10:29 AM

i just completed a training for a client with CogVLM2 captions

Furkan Gözükara SECourses•7/29/24, 10:30 AM

and results looking stunning

Furkan Gözükara SECourses•7/29/24, 10:30 AM

still comparing with raw prompt

Furkan Gözükara SECourses•7/29/24, 2:49 PM

i just did a new training for 40 images

Furkan Gözükara SECourses•7/29/24, 2:50 PM

after 20k steps it started to became overtrained

Furkan Gözükara SECourses•7/29/24, 2:50 PM

reg images used

Furkan Gözükara SECourses•7/29/24, 2:52 PM

raw captions still provides better likeliness

Furkan Gözükara SECourses•7/29/24, 2:58 PM

i feel like 200 epoch yielded best

Furkan Gözükara SECourses•7/29/24, 2:58 PM

200 x 40 x 2 = 16000 steps

pat._.fenis•7/29/24, 3:12 PM

What's the most recent koyha install/ guide?

There are so many Patreon posts that I don't use that I can no longer keep up with the ones I might use.
Is there some way to sort them that I am missing?

Ppat._.fenis What's the most recent koyha install/ guide? There are so many Patreon posts t...

Furkan Gözükara SECourses•7/29/24, 3:35 PM

where you want to install on windows?

pat._.fenis•7/29/24, 3:43 PM

yes windows 10

FFurkan Gözükara SECourses it could be also due to your training config

dsienra•7/29/24, 5:19 PM

update, with 25 epochs 3 repetitions and the TE learning rate at the same learning rate as the unet, the bleeding I think is fixed, I'm no so sure yet because is undertrained, I will resume the training for 25 more ephocs but the results are promising

Ppat._.fenis yes windows 10

Furkan Gözükara SECourses•7/29/24, 5:49 PM

kohya gui has auto install bat file

Furkan Gözükara SECourses•7/29/24, 5:49 PM

itself

Furkan Gözükara SECourses•7/29/24, 5:49 PM

you only need to have python 3.10

Furkan Gözükara SECourses•7/29/24, 5:50 PM

up to date config is here : https://www.patreon.com/posts/very-best-for-of-89213064

Patreon

The Very Best Workflow For SDXL DreamBooth / Full Fine Tuning - Res...

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

Ddsienra update, with 25 epochs 3 repetitions and the TE learning rate at the same learni...

Furkan Gözükara SECourses•7/29/24, 5:50 PM

great

Furkan Gözükara SECourses•7/29/24, 10:05 PM

i plan to fine tune sd3 with huge dataset

Furkan Gözükara SECourses•7/29/24, 10:06 PM

non-commercial use free

Furkan Gözükara SECourses•7/29/24, 10:06 PM

individual commercial 50$ and company 500$ . what you think

Furkan Gözükara SECourses•7/29/24, 10:06 PM

starting with 100k then 500k then 1m and so on dataset size

Furkan Gözükara SECourses•7/29/24, 10:06 PM

but first i need to find best hyper parameters :d

Furkan Gözükara SECourses•7/29/24, 10:07 PM

so many things to do so little time

FFurkan Gözükara SECourses because of the base models

Narva•7/29/24, 10:19 PM

Any tip for train some style checkpoint for asian model? I have like around 8k of hq pics of asian people 50% woman / 50% male I already did the caption thing using LLava, so using ur config can be correctly?

magical•7/29/24, 10:27 PM

@Dr. Furkan Gözükara hello Dr. are you planning to make a video about sd3 training with kohya? can we use the same settings we used for sdxl training?

FFurkan Gözükara SECourses individual commercial 50$ and company 500$ . what you think

Haashim•7/29/24, 10:39 PM

Could work, would really just depend on how well the model performs.

Have you had any success with a trial training?

NNarva Any tip for train some style checkpoint for asian model? I have like around 8k o...

Furkan Gözükara SECourses•7/29/24, 10:49 PM

use our config and llava or cogvlm caption them

Mmagical @Dr. Furkan Gözükara hello Dr. are you planning to make a video about sd3 traini...

Furkan Gözükara SECourses•7/29/24, 10:49 PM

yes hopefully once i found good config and workflow

HHaashim Could work, would really just depend on how well the model performs. Have you...

Furkan Gözükara SECourses•7/29/24, 10:49 PM

not yet

Furkan Gözükara SECourses•7/29/24, 10:49 PM

first i need to research

Furkan Gözükara SECourses•7/30/24, 12:03 AM

2GB x/y/z plot

highjourney•7/30/24, 2:53 PM

Does anybody know how to train a LoRA for inpainting base models?

Currently dealing with the error below with kohya_ss.

ValueError: Trying to set a tensor of shape torch.Size([320, 9, 3, 3]) in "weight" (which has shape torch.Size([320, 4, 3, 3])), this look incorrect.