Software Engineering Courses (SECourses)•3y ago

Hi @Furkan Gözükara SECourses. Automatic1111 has had many merges since the end of last month, I see

Hi @Furkan Gözükara SECourses. Automatic1111 has had many merges since the end of last month, I see that torch 2.0 and possibly a newer version of xformers is already in use. Is this a problem when using runpod? I used to follow your tutorial https://www.youtube.com/watch?v=zA4LksIVas8 along with this file: https://gist.github.com/FurkanGozukara/fdfb1c24a38509c8ec0184951ba12abf
but now it may be outdated.

YouTubeSECourses

RunPod Fix For DreamBooth & xFormers - How To Use Automatic1111 Web...

RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. How to install latest version of Automatic1111 Web UI and #DreamBooth on #RunPod. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of #StableDiffusi...

RunPod Fix For DreamBooth & xFormers - How To Use Automatic1111 Web...

Gist

How to run and install Automatic1111 Web UI and DreamBooth on RunPo...

How to run and install Automatic1111 Web UI and DreamBooth on RunPod with latest xformers - runpod xformers.txt

AAusent315 Hi @Furkan Gözükara SECourses. Automatic1111 has had many merges since the end o...

Furkan Gözükara SECourses•5/4/23, 2:40 PM

for runpod

Furkan Gözükara SECourses•5/4/23, 2:40 PM

torch 2 was not working

Furkan Gözükara SECourses•5/4/23, 2:40 PM

let me test again

Furkan Gözükara SECourses•5/4/23, 2:40 PM

also how is looking thumbnail

Furkan Gözükara SECourses•5/4/23, 2:40 PM

weilu•5/4/23, 2:48 PM

@Furkan Gözükara SECourses
Have you tried finetuning on DeepFloyd IF?

FFurkan Gözükara SECourses torch 2 was not working

Ausent315OP•5/4/23, 3:00 PM

For what it's worth, I just tested it and it does work (I haven't tried dreambooth, only automatic 1111). I upgraded to the latest version

git pull

git pull

, used the --reinstall-torch--reinstall-torch command to force it to install 2.0.0, installed the latest version of xformers directly with pip install xformerspip install xformers and it works correctly.
Here are the parameters: python: 3.10.10 • torch: 2.0.0+cu117 • xformers: 0.0.19 • gradio: 3.28.1 • commit: 5ab7f213 • checkpoint: 4c86efd062

Zono50•5/4/23, 3:19 PM

yeah i'm getting an error when ruunning find face

Zono50•5/4/23, 3:19 PM

says valueerror: face could not be detected, please confirm that the picture is a face photo or consider to set enforce_detection param fo false

Zono50•5/4/23, 3:20 PM

Zono50•5/4/23, 3:20 PM

Traceback (most recent call last):
File "/home/zono50/Scripts/findBestImages.py", line 35, in <module>
detected_face = DeepFace.detectFace(org_image_path, detector_backend='opencv')
File "/home/zono50/anaconda3/envs/findbestimages/lib/python3.10/site-packages/deepface/DeepFace.py", line 820, in detectFace
face_objs = extract_faces(
File "/home/zono50/anaconda3/envs/findbestimages/lib/python3.10/site-packages/deepface/DeepFace.py", line 787, in extract_faces
img_objs = functions.extract_faces(
File "/home/zono50/anaconda3/envs/findbestimages/lib/python3.10/site-packages/deepface/commons/functions.py", line 115, in extract_faces
raise ValueError(
ValueError: Face could not be detected. Please confirm that the picture is a face photo or consider to set enforce_detection param to False.

Wweilu @Furkan Gözükara SECourses Have you tried finetuning on DeepFloyd IF?

Furkan Gözükara SECourses•5/4/23, 3:51 PM

nope

AAusent315 For what it's worth, I just tested it and it does work (I haven't tried dreamboo...

Furkan Gözükara SECourses•5/4/23, 3:51 PM

ye it wasnt working for dreambooth on unix

ZZono50 Click to see attachment

Furkan Gözükara SECourses•5/4/23, 3:51 PM

it can happen in some faces

Furkan Gözükara SECourses•5/4/23, 3:52 PM

if it happens in your training images

Furkan Gözükara SECourses•5/4/23, 3:52 PM

remove them

Furkan Gözükara SECourses•5/4/23, 3:52 PM

from folder

Furkan Gözükara SECourses•5/4/23, 3:52 PM

your org folder must have good pictures

Furkan Gözükara SECourses•5/4/23, 3:52 PM

that face can be deteceted

Furkan Gözükara SECourses•5/4/23, 3:52 PM

but i will update script

Furkan Gözükara SECourses•5/4/23, 3:52 PM

for auto removal good idea

Furkan Gözükara SECourses•5/4/23, 3:54 PM

working on script update

TheRustingDeadAdmin•5/4/23, 4:00 PM

Can stable diffusion train on bearded men? It cant seem to get good results on my own face and i think thats why

TheRustingDeadAdmin•5/4/23, 4:03 PM

i mean big beards, mine is a full foot long probably more

Em•5/4/23, 4:15 PM

Hello, need some help on copying some SD files from one pod to another, watched the tutorial on youtube but it errors out. Not sure if this is the right channel to ask, if not, apologies in advance.

Zono50•5/4/23, 4:17 PM

yeah nothing i seem to do will get this womans face properly

ZZono50 yeah i'm getting an error when ruunning find face

Furkan Gözükara SECourses•5/4/23, 4:20 PM

updated script you can redownload

TTheRustingDeadAdmin i mean big beards, mine is a full foot long probably more

Furkan Gözükara SECourses•5/4/23, 4:21 PM

i think yes why not

TheRustingDeadAdmin•5/4/23, 4:23 PM

probably my dataset then, I have a feeling the transparency of beards can cause issue

Zono50•5/4/23, 4:37 PM

doc my wife is going to divorce me if i don't make a good model of her

Zono50•5/4/23, 4:37 PM

no pressure

ivaxsirc -4090•5/4/23, 4:41 PM

I have entered discord, because I was a teacher, and they are the best tutorials for the use and installation of stable diffusion that I have seen of all the ones on youtube, not only for the explanations of how to do the things, but there is an addition where You explain, for example, in dreambooth the parameters for which they are used in detail, then you also apply different resolutions and compare the results.
It is let's say a very scientific methodology, congratulations

Zono50•5/4/23, 5:03 PM

is having 200 images in my dataset using 5 class reg images the same as using 10 images in my dataset using 5 class reg images

Zono50•5/4/23, 5:03 PM

like how do i get it to learn every intricate detail of someones face

Iivaxsirc -4090 I have entered discord, because I was a teacher, and they are the best tutorials...

Furkan Gözükara SECourses•5/4/23, 5:05 PM

thank you so much

Furkan Gözükara SECourses•5/4/23, 5:05 PM

yes I am also a teacher. Assistant professor in a private university in Turkey

Furkan Gözükara SECourses•5/4/23, 5:06 PM

new video almost ready. waiting youtube to synch subtitles hopefully

Zono50•5/4/23, 5:24 PM

I did a search on stable diffusion, saw an article on linkedin, guess whose it was

Zono50•5/4/23, 5:34 PM

is the only point of the num class instancers per image or regularization images is to preserve the class?

Zono50•5/4/23, 5:34 PM

so if it overfits, if i type woman, it will only show the woman in my dataset, and no other women?

Zono50•5/4/23, 5:38 PM

can you explain doc?

Zono50•5/4/23, 5:43 PM

@Furkan Gözükara SECourses

ZZono50 is the only point of the num class instancers per image or regularization images...

Furkan Gözükara SECourses•5/4/23, 5:53 PM

not only that

Furkan Gözükara SECourses•5/4/23, 5:53 PM

it also prevents overtraining

ZZono50 so if it overfits, if i type woman, it will only show the woman in my dataset, a...

Furkan Gözükara SECourses•5/4/23, 5:54 PM

if you train very long, even when you dont type anything you get your trained subject

Furkan Gözükara SECourses•5/4/23, 5:54 PM

because unet becomes cooked

Zono50•5/4/23, 5:57 PM

overtraining is when the images start looking warped?

ZZono50 overtraining is when the images start looking warped?

Furkan Gözükara SECourses•5/4/23, 5:58 PM

usually

TheRustingDeadAdmin•5/4/23, 6:00 PM

I have a question, when you use the "ohwx man" and the <lora> to generate an image, are you still increasing the weights using ctrl+up e.g. (ohwx man:1.5) or are you keeping it at 1.1?
2nd question; Do we need both [the lora and the ohwx man] everytime we want to call our model?

TTheRustingDeadAdmin I have a question, when you use the "ohwx man" and the <lora> to generate an ima...

Furkan Gözükara SECourses•5/4/23, 6:01 PM

i usually test this because sometimes 1.1 or 1.2 works better

Hi @Furkan Gözükara SECourses. Automatic1111 has had many merges since the end of last month, I see

Similar Threads

Similar Threads

Similar Threads