Software Engineering Courses (SECourses)•2y ago

I thought I had read that the Lora extraction was deprecated in the backend code, but still in the U

I thought I had read that the Lora extraction was deprecated in the backend code, but still in the UI. Are there any special settings that should be used for a successful extraction?

DDJRich I thought I had read that the Lora extraction was deprecated in the backend code...

Furkan Gözükara SECourses•12/29/23, 11:49 PM

it works great in kohya gui

Furkan Gözükara SECourses•12/29/23, 11:49 PM

i just tested

Furkan Gözükara SECourses•12/29/23, 11:49 PM

try different ranks

Furkan Gözükara SECourses•12/29/23, 11:49 PM

such as

Furkan Gözükara SECourses•12/29/23, 11:49 PM

128x128

Furkan Gözükara SECourses•12/29/23, 11:49 PM

192x192

Furkan Gözükara SECourses•12/29/23, 11:49 PM

256x256

Furkan Gözükara SECourses•12/29/23, 11:49 PM

whichever you like most

DJRichOP•12/29/23, 11:50 PM

Is there a json on Patreon for 16GB VRAM local Dreambooth training?

FFurkan Gözükara SECourses whichever you like most

DJRichOP•12/29/23, 11:55 PM

Or should I just use the Lora preset in the DreamBooth tab?

DDJRich Is there a json on Patreon for 16GB VRAM local Dreambooth training?

Furkan Gözükara SECourses•12/29/23, 11:56 PM

yes

Furkan Gözükara SECourses•12/29/23, 11:56 PM

we have here

Furkan Gözükara SECourses•12/29/23, 11:56 PM

https://www.patreon.com/posts/very-best-for-of-89213064

Furkan Gözükara SECourses•12/29/23, 11:56 PM

16GB_TextEncoder.json

Furkan Gözükara SECourses•12/29/23, 11:57 PM

i did 100+ testing and i believe my findings much better than those presets

FFurkan Gözükara SECourses have you tested my auto installer?

Digital [Starburst]•12/30/23, 12:58 AM

Yes I used the autoinstaller. My GPU is a 4070 ti

FFurkan Gözükara SECourses i did 100+ testing and i believe my findings much better than those presets

DJRichOP•12/30/23, 1:04 AM

Thank you. I ended up using your settings from the Lora training video you made. The issue is that the video is five months old and I know how quickly the technology is advancing. I will try the DreamBooth preset and let you know how it goes.

DDigital [Starburst]Yes I used the autoinstaller. My GPU is a 4070 ti

Furkan Gözükara SECourses•12/30/23, 1:21 AM

how much vram it has?

DDJRich Thank you. I ended up using your settings from the Lora training video you made....

Furkan Gözükara SECourses•12/30/23, 1:21 AM

i did very extensive serach for dreambooth

Furkan Gözükara SECourses•12/30/23, 1:21 AM

lora was not that much extensive

Furkan Gözükara SECourses•12/30/23, 1:21 AM

also dreambooth far superior than lora i compared

FFurkan Gözükara SECourses i did 100+ testing and i believe my findings much better than those presets

DJRichOP•12/30/23, 1:23 AM

By the way, I think I figured out the issue. I was using too many regularization images. I am doing a dreambooth test now using your text encoder 16GB preset with no modifications. I'll post my results when it completes. Thanks again

FFurkan Gözükara SECourses also dreambooth far superior than lora i compared

Haashim•12/30/23, 2:26 AM

I need to test this. Will check out recent vids

FFurkan Gözükara SECourses hello. are you installing on activated venv?

banfi6885•12/30/23, 5:55 AM

Hello, I finally got it working. Also, have you trained a style on the "dreamshaper xl turbo" model? If yes, did you use colab or runpod with kohya? Thank you

Deleted User•12/30/23, 8:08 AM

Hi @Dr. Furkan Gözükara My DB test finished a short time ago and I am still seeing similar results. I ran x,y,z plot tests and while facial features were similar, some prominent characteristics such as the subject's long hair and blue eyes were not present. One example prompt I used had instagram selfie at the start of the prompt and even the last epoch's model turned the subject into a young, muscular man when the subject is in his 60's. I was able to see slightly better results by using (Instance prompt Class prompt:2), but I feel that shouldn't be necessary if I am training correctly.

I've included a screenshot of my dataset preparation page if it helps. I am also using the 16GB_TextEncoder.json provided with no changes at all other than modifying the output model name.

Deleted User•12/30/23, 10:06 AM

DDeleted User Click to see attachment

Deleted User•12/30/23, 10:08 AM

Please guide me i have 15 images repeat 150 + regularisation images batch size 2 total steps is 4500, on how much steps i need to save every n steps for best quality checkpoints

Deleted User•12/30/23, 10:28 AM

15 * 30 * 2 + 1=901 is this ok for checkpoints to save

DDeleted User Hi @Dr. Furkan Gözükara My DB test finished a short time ago and I am still see...

DJRichOP•12/30/23, 12:42 PM

I ran one more test overnight. The only thing I changed is that I pointed the regularization image folder to the one that has the 5200 images you compiled whereas before I was calculating the images needed for the number of repeats I was doing for the subject images and creating a folder with that total number. I'm not sure if that would make a difference, but it sure seems to. The likeness is much better now. The only challenge which I don't know the resolution for yet is that when putting additional names in the prompt such as the artist's style, it does cause issues with the likeness. I am not sure if perhaps doing more epochs is the answer or something else.

andysingal•12/30/23, 1:30 PM

My New Years Article, I hope everyone enjoys it: Shakespeare’s Digital Odyssey: Poesy in the Age of AI https://medium.com/ai-advances/shakespeares-digital-odyssey-poesy-in-the-age-of-ai-773e4bdb6054

Medium

Shakespeare’s Digital Odyssey: Poesy in the Age of AI

Ankush k Singal

Bbanfi6885 Hello, I finally got it working. Also, have you trained a style on the "dreamsha...

Furkan Gözükara SECourses•12/30/23, 2:48 PM

i didnt

DDeleted User Hi @Dr. Furkan Gözükara My DB test finished a short time ago and I am still see...

Furkan Gözükara SECourses•12/30/23, 2:49 PM

how many epoch did you train with 20 repeat?

FFurkan Gözükara SECourses how many epoch did you train with 20 repeat?

DJRichOP•12/30/23, 2:50 PM

The preset fille you provided defaulted to 8 epochs. It ended up being roughly 9600 steps in total.

DDJRich The preset fille you provided defaulted to 8 epochs. It ended up being roughly 9...

Furkan Gözükara SECourses•12/30/23, 2:50 PM

i see

Furkan Gözükara SECourses•12/30/23, 2:50 PM

ye 9600 steps is big

Furkan Gözükara SECourses•12/30/23, 2:50 PM

it should have memorized already

DJRichOP•12/30/23, 2:53 PM

Would increasing the number of repeating from 20 to 25 improve the likeness without having to do this? (nckcv man:1.4), etc?

JMPA•12/30/23, 2:58 PM

Hi guys, I put my logo on controlnet on Inpaint section, but I select de paint part it should be put, but it doesn’t resize to that area only

JMPA•12/30/23, 2:58 PM

This is the area I mark

JMPA•12/30/23, 2:59 PM

And this is how the logo shows

JMPA•12/30/23, 2:59 PM

And this is the whole logo text… I don’t understand how to put it only on the marked area

AIGambino•12/30/23, 3:54 PM

Any idea why sometimes itll only export the same image over and over despite it being a png sequence I'm trying to render out?
All controlnets are selecting the batch tab with correct directory

AIGambino•12/30/23, 4:00 PM

first image rendered and 7th image rendered.. same image for some reason

Vlad•12/30/23, 5:06 PM

Hello @Dr. Furkan Gözükara can I please ask you 2 questions?
1. is it better to train a LORA on the base SD1.5 model instead of a realistic checkpoint? (for a realistic character)
2. can I train a LORA on SDXL if my training images are 768*768?

FFurkan Gözükara SECourses how much vram it has?

Digital [Starburst]•12/30/23, 5:08 PM

12 GB

Digital [Starburst]•12/30/23, 5:10 PM

I know you said for Dreambooth 16 GB is required, but I believe I read in the instructions that using Kohya I could get by with 12 GB (although clearly I'm pushing my luck). However, the install uses Automatic1111 and to use it I require TensorRT to run it. So I feel like a fish out of water right now.

Digital [Starburst]•12/30/23, 5:11 PM

It's running out of memory when it trys to build the engine. With each step I get the Out of Memory error and finally it crashes altogether and says Error in big red letters in the UI.

Digital [Starburst]•12/30/23, 5:12 PM

I tried building the 512 to 768 dynamic engine alone which also failed. I wanted to build the Default engine.

Digital [Starburst]•12/30/23, 5:13 PM

My CUDA and CUDNN version is 11.8. When I swapped that out for 12.3 it wouldn't work at all, so clearly that is wrong.

Digital [Starburst]•12/30/23, 5:15 PM

I'm also in the process of taking this course: https://www.udemy.com/course/master-ai-image-generation-using-stable-diffusion/ which teaches you to use SD completely from code with no UI. He claims that doing so gives you more control over certain aspects of image generation. Truthfully I think learning both aspects of SD are important. The use of UIs and the use of code because learning to code it teaches you more about how it works.

I thought I had read that the Lora extraction was deprecated in the backend code, but still in the U

Similar Threads

Similar Threads

Similar Threads