Software Engineering Courses (SECourses)•3y ago

Huge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison U-NET is always trained

Huge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison

U-NET is always trained

PNG info (prompts) are in alts

All images are 1024x1024 so download full sizes

https://twitter.com/GozukaraFurkan/status/1711345323900100643

Benjamin•10/9/23, 11:40 AM

v23.10.0 adetailer got released. Just noticed because now in the preview of the image a red box appears around the detected face

FFurkan Gözükara SECourses Huge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison U-N...

Benjamin•10/9/23, 11:41 AM

good job, from a first glance it does not look like it is changing that much? what is your impression

BBenjamin v23.10.0 adetailer got released. Just noticed because now in the preview of the ...

Furkan Gözükara SECoursesOP•10/9/23, 12:44 PM

2023-10-07

- v23.10.0
- 허깅페이스 모델을 다운로드 실패했을 때, 계속 다운로드를 시도하지 않음
- img2img에서 img2img단계를 건너뛰는 기능 추가
- live preview에서 감지 단계를 보여줌 (PR #352)

BBenjamin good job, from a first glance it does not look like it is changing that much? wh...

Furkan Gözükara SECoursesOP•10/9/23, 12:45 PM

i think it still adds some more extra details

Furkan Gözükara SECoursesOP•10/9/23, 12:45 PM

so 2e-6 looks like nice. 3e-6 also nice but a little bit starting to cook

FFurkan Gözükara SECourses i think it still adds some more extra details

Benjamin•10/9/23, 12:47 PM

how did it influence the training time?

Benjamin•10/9/23, 12:48 PM

ah you already wrote in patreon it is slower

BBenjamin ah you already wrote in patreon it is slower

Furkan Gözükara SECoursesOP•10/9/23, 12:54 PM

ye a little bit more slower

Furkan Gözükara SECoursesOP•10/9/23, 12:54 PM

like 1.5 second / it to 1.75 second / it

Stefanoid9343•10/9/23, 1:11 PM

Thank you so much to @Dr. Furkan Gözükara and @Sandstormleader after several hours of training and solving some errors the model came as I expected this is definitely a game changer process

FFurkan Gözükara SECourses i never got good results with Prodigy

mikemenders•10/9/23, 1:14 PM

I have now made a perfect Lora with Prodigy, but you have to use Clip Skip 2 when generating and then it will be flexible.

Mmikemenders I have now made a perfect Lora with Prodigy, but you have to use Clip Skip 2 whe...

Furkan Gözükara SECoursesOP•10/9/23, 1:20 PM

interesting never tested clip skip 2. this is sd 1.5?

SStefanoid9343 Thank you so much to @Dr. Furkan Gözükara and @Sandstormleader after several hou...

Furkan Gözükara SECoursesOP•10/9/23, 1:20 PM

awesome

FFurkan Gözükara SECourses interesting never tested clip skip 2. this is sd 1.5?

mikemenders•10/9/23, 1:23 PM

Yes, I trained with RealisticVision 5.1, and I generated the images in the same model.

mikemenders•10/9/23, 1:24 PM

If I generated on clip skip 1, the model was inflexible, but on clip skip 2 everything was fine.
And I thought right: I trained it with clip skip 2 and then clip skip 1 can be generated well.

Metaphysix•10/9/23, 1:31 PM

So now the SDXL training works properly with TE, but what's up with 1.5 dreambooth?

rafraf•10/9/23, 1:51 PM

what is the best/fastest gpu deal on Runpod to do inference on Auto1111? A100 or A6000? I find A6000 very slow, slower than 4900...

MMetaphysix So now the SDXL training works properly with TE, but what's up with 1.5 dreamboo...

Furkan Gözükara SECoursesOP•10/9/23, 1:51 PM

yes but using kohya

Furkan Gözükara SECoursesOP•10/9/23, 1:52 PM

i think sd 1.5 dreambooth also should work but need good hyper parameter testing. also kohya missing EMA

Rrafraf what is the best/fastest gpu deal on Runpod to do inference on Auto1111? A100 or...

Furkan Gözükara SECoursesOP•10/9/23, 1:52 PM

4090 could be fastest with price ratio

FFurkan Gözükara SECourses 4090 could be fastest with price ratio

rafraf•10/9/23, 1:56 PM

ok, yea, that's what i thought. a6000 takes twice as long to infer

EEyeSpyBekoAI Hey gang, does anyone have experience on downloading safetensor files from runpo...

Knechtruprecht•10/9/23, 2:09 PM

use "runpodctl" to transfer big Data between your desktop and runpod https://github.com/runpod/runpodctl

GitHub

GitHub - runpod/runpodctl: 🧰 | RunPod CLI for pod management

| RunPod CLI for pod management. Contribute to runpod/runpodctl development by creating an account on GitHub.

GitHub - runpod/runpodctl: 🧰 | RunPod CLI for pod management

thesilvermoth•10/9/23, 2:13 PM

I tested 4090 with Kohya on Windows 11 and it was 1.89 it. On Windows 10 it is 1.06 it. Seems like there is a huge driver issue with W11

FFurkan Gözükara SECourses i think sd 1.5 dreambooth also should work but need good hyper parameter testing...

Metaphysix•10/9/23, 2:13 PM

I'm speaking about kohyaSS 1.5 dreambooth training. Can I train SD1.5 with TE?

KKnechtruprecht use "runpodctl" to transfer big Data between your desktop and runpod https://git...

Furkan Gözükara SECoursesOP•10/9/23, 2:14 PM

uploading hugging face is even faster

Tthesilvermoth I tested 4090 with Kohya on Windows 11 and it was 1.89 it. On Windows 10 it is 1...

Furkan Gözükara SECoursesOP•10/9/23, 2:14 PM

100%

Furkan Gözükara SECoursesOP•10/9/23, 2:14 PM

it works too slow on windows

Furkan Gözükara SECoursesOP•10/9/23, 2:15 PM

driver issue

MMetaphysix I'm speaking about kohyaSS 1.5 dreambooth training. Can I train SD1.5 with TE?

Furkan Gözükara SECoursesOP•10/9/23, 2:15 PM

yes we found you can

Furkan Gözükara SECoursesOP•10/9/23, 2:15 PM

but it is inferior

thesilvermoth•10/9/23, 2:15 PM

But I'm using it on Windows 10 and it works really fast. Seems like Windows 11 is the problem

Tthesilvermoth But I'm using it on Windows 10 and it works really fast. Seems like Windows 11 i...

Furkan Gözükara SECoursesOP•10/9/23, 2:18 PM

this is weird

Furkan Gözükara SECoursesOP•10/9/23, 2:18 PM

so windows 11 is the culprit?

Furkan Gözükara SECoursesOP•10/9/23, 2:18 PM

i am still using 10

FFurkan Gözükara SECourses so windows 11 is the culprit?

thesilvermoth•10/9/23, 2:20 PM

Yes, look. This is on Windows 10. I tested the same hardware with Windows 11 on a partition of the same SSD and I got 1.89s / it

thesilvermoth•10/9/23, 2:22 PM

And also tested Automatic 1111 on Windows 11 and the generation time was terrible

thesilvermoth•10/9/23, 2:22 PM

it's not worth updating the OS

Tthesilvermoth Yes, look. This is on Windows 10. I tested the same hardware with Windows 11 on ...

Furkan Gözükara SECoursesOP•10/9/23, 2:23 PM

huge difference

Tthesilvermoth it's not worth updating the OS

Furkan Gözükara SECoursesOP•10/9/23, 2:23 PM

that is why i am still on 10

Furkan Gözükara SECoursesOP•10/9/23, 2:23 PM

it works super good

Sandstormleader•10/9/23, 2:45 PM

is there anything like runway gen2 inside of stable diffusion?

Sandstormleader•10/9/23, 2:45 PM

@Dr. Furkan Gözükara

LEGACIES•10/9/23, 3:20 PM

If i train my images 1024 x 1024 on dreambooth, will my model have better quality than generated with 512 x 512? Does that also mean that i will have to generate my images 1024 x 1024 with my model?

Sandstormleader•10/9/23, 3:21 PM

i had better luck with 768 but was probably my base model not being great for 1024

LEGACIES•10/9/23, 3:23 PM

Can i train 768 on a 12gb gpu?

LEGACIES•10/9/23, 3:24 PM

For now i am doing training with 512px

mtnmecca•10/9/23, 3:48 PM

rtx 3090 vs 4090 24gb? extra cuda cores worth the extra cost?

FFurkan Gözükara SECourses huge difference

mtnmecca•10/9/23, 4:12 PM

thanks for info. just upgraded to 11 and moving back now..

FFurkan Gözükara SECourses you need a specific version

kesenteboregi•10/9/23, 4:23 PM

stable diffusion or dreambooth ?

Stefanoid9343•10/9/23, 6:28 PM

@Dr. Furkan Gözükara By chance are you planning or you already have a tutorial of the importance and explanation of the image regularization data? how this works also with the caption description?

Huge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison U-NET is always trained

2023-10-07

Similar Threads

Similar Threads

2023-10-07

Similar Threads