Software Engineering Courses (SECourses)•3y ago

@Dr. Furkan Gözükara, what would be the best way to train a person in different ages? Is it a good i

@Dr. Furkan Gözükara, what would be the best way to train a person in different ages? Is it a good idea? Should I use captions?

Furkan Gözükara SECourses•10/24/23, 8:58 PM

also use exact versions

MMarcelo @Dr. Furkan Gözükara, what would be the best way to train a person in different ...

Furkan Gözükara SECourses•10/24/23, 8:59 PM

do you have different age photos?

FFurkan Gözükara SECourses also use exact versions

Crazy dream•10/24/23, 8:59 PM

Thank you for your answer! even if I don't have any issue with the current ones ? (I still don't know if it will work but in the video, I understood you advice us to don't use exact version if it works)

MarceloOP•10/24/23, 9:00 PM

Yes, in this dataset I have the same person, but in different epochs, young adult, adult....

MarceloOP•10/24/23, 9:01 PM

For example, Christian Bale now and 10 years ago

CCrazy dream Thank you for your answer! even if I don't have any issue with the current ones ...

Furkan Gözükara SECourses•10/24/23, 9:07 PM

use exact versions i used in tutorial

FFurkan Gözükara SECourses use exact versions i used in tutorial

Crazy dream•10/24/23, 9:07 PM

ok, thank you very much

_Captain_•10/24/23, 9:11 PM

@Dr. Furkan Gözükara thanks for your research! I'm one of your patreons now

I have a question about text encoder training and captions.
(Using Koyha Dreambooth training / 3090ti - your sdxl dreambooth workflow works really well!)

To my understanding SDXL text encoder is based on CLIP ViT-L & OpenClip ViT-bigG.
So the model understands only words the text encoder contains. So as long I'm training things the text encoder knows, everything is fine (+ you don't have to train the text encoder).

But let's take a special martial arts pose for example. Some of them have uniqe names and using your "SOTA (The Very Best) Image Captioning..." tutorial, it was easy to prove that the prompt won't contain the correct names for the pose.

When training the model using the dreambooth method, every image has it's textfile with
instance, class, image description -> for example: I want to train a chinese kung fu stance named "pubu" and the model has no knowledge of it ->
pubu, man, in a training hall,

NOW! If I train without the textencoder:
What impact have the captions? Is "man, in a training hall" a guide for the noise generation?
Will the new images enhance the previous knowledge of a man in a training hall?

I assume "pubu" has no impact on the training at all? I have no idea how the unet is working with these captions.
Or does the model somehow learn about "pubu"?

Is this the point where you have to train the text encoder? I tried it once with the same settings and the model (text encoder) was waaay overtrained. EDIT: "just found your settings for training with text encoder, will try that!"

Summery Question:
If I use captions the text encoder doesn't know -> I assume I cannot use them later for generating images.
(So I have to train the text encoder too)

Sorry for the long question, don't know how to shorten all this complexe stuff xD

FFurkan Gözükara SECourses use exact versions i used in tutorial

Crazy dream•10/24/23, 9:11 PM

last question sorry : I have to use exact version used in this tutorial https://www.youtube.com/watch?v=g0wXIcRhkJk right ? (not this one https://www.youtube.com/watch?v=c_S2kFAefTQ)

YouTubeSECourses

The END of Photography - Use AI to Make Your Own Studio Photos, FRE...

Dreambooth is the best training method for Stable Diffusion. In this tutorial, I show how to install the Dreambooth extension of Automatic1111 Web UI from scratch. Additionally, I demonstrate my months of work on the realism workflow, which enables you to produce studio-quality images of yourself through #Dreambooth training. Furthermore, I shar...

YouTubeSECourses

How To Install DreamBooth & Automatic1111 On RunPod & Latest Librar...

I have shown how to install latest version of Automatic1111 Web UI for Stable Diffusion and DreamBooth extension of Auto1111 on RunPod in this video. Moreover, I show how to upgrade to latest Cuda, Torch and cuDNN DLL files. With these upgrades the image generation speed literally doubles.

GitHub Readme File

https://github.com/FurkanGozukara...

How To Install DreamBooth & Automatic1111 On RunPod & Latest Librar...

Zono50•10/24/23, 9:23 PM

trying to launch kohya, and getting scipy errors, even after installing both with pip and then conda

Mmikemenders Sorry, there's no --lowvram-sdxl switch, so it's still slow because you're gener...

vKorby•10/24/23, 9:37 PM

For some reason now its running pretty quick lol, with 1024 too

akellio•10/24/23, 9:38 PM

Regularization images in runpod training need to be loaded manually?

akellio•10/24/23, 10:05 PM

outfigured out

Crazy dream•10/24/23, 10:20 PM

I have this error :
"/workspace/venv/lib/python3.10/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: '/workspace/venv/lib/python3.10/site-packages/torchvision/image.so: undefined symbol: _ZN3c104cuda9SetDeviceEi'If you don't plan on using image functionality from torchvision.iotorchvision.io, you can ignore this warning. Otherwise, there might be something wrong with your environment. Did you have libjpeglibjpeg or libpnglibpng installed before building torchvisiontorchvision from source?
[...]
OSError: /workspace/venv/lib/python3.10/site-packages/torchaudio/lib/libtorchaudio.so: undefined symbol: _ZN3c104cuda9SetDeviceEi
Relauncher: Process is ending. Relaunching in 2s..." However I am following the tutorial https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Tutorials/How-To-Install-DreamBooth-Extension-On-RunPod.md step by step. Somebody would know why ?

GitHub

Stable-Diffusion/Tutorials/How-To-Install-DreamBooth-Extension-On-R...

Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Automatic1111 Web UI, DeepFake, Deep Fakes, TTS, Animation, Text To Video, Tutorials, Guides, Lectures, Courses, ComfyUI, Google Colab, R...

MMarcelo For example, Christian Bale now and 10 years ago

mikemenders•10/24/23, 10:30 PM

Check Age sliders on civitai. These special loras:
https://civitai.com/search/models?sortBy=models_v2&query=age%20slider

CCrazy dream I have this error : "/workspace/venv/lib/python3.10/site-packages/torchvision/i...

Crazy dream•10/24/23, 10:31 PM

I have the following errors "root@xxx:/workspace/stable-diffusion-webui# fuser -k 3000/tcp
cd /workspace/stable-diffusion-webui
python relauncher.py
Cannot stat file /proc/43/fd/0: Permission denied
Cannot stat file /proc/43/fd/1: Permission denied
Cannot stat file /proc/43/fd/2: Permission denied
Cannot stat file /proc/43/fd/4: Permission denied
Cannot stat file /proc/43/fd/5: Permission denied
Cannot stat file /proc/43/fd/6: Permission denied
Cannot stat file /proc/43/fd/7: Permission denied
Cannot stat file /proc/43/fd/8: Permission denied
Cannot stat file /proc/43/fd/9: Permission denied
Cannot stat file /proc/43/fd/10: Permission denied
Cannot stat file /proc/43/fd/11: Permission denied
Cannot stat file /proc/43/fd/12: Permission denied
Cannot stat file /proc/43/fd/13: Permission denied
3000/tcp: 91
Relauncher: Launching..." on the tutorial, there isn't these "pemission denied". Could the problem come from here ?

JJas Thanks for your recommendations, with the labelling, I am a bit uncertain how to...

mikemenders•10/24/23, 10:32 PM

If you want to train something like this, it's better to create a generator, so I wouldn't tag it, but train it like the face, with a folder name, without any reg images, for example: 10_mystyle style. And then SD will interpret what it "sees", and you will only have to specify the colors in the prompt.

mikemenders•10/24/23, 10:33 PM

With tagging you are fine-tuning, that is, you are retraining existing tags. Since this doesn't have a label yet, it's more of a style and a new tag.

Crazy dream•10/24/23, 10:43 PM

Do somebody would know any forum/discord for very newbie please ?

ZZono50 trying to launch kohya, and getting scipy errors, even after installing both wit...

Furkan Gözükara SECourses•10/24/23, 10:59 PM

use updated installer from patreon

Aakellio Regularization images in runpod training need to be loaded manually?

Furkan Gözükara SECourses•10/24/23, 10:59 PM

yes

CCrazy dream I have this error : "/workspace/venv/lib/python3.10/site-packages/torchvision/i...

Furkan Gözükara SECourses•10/24/23, 11:00 PM

use auto installer from patreon

CCrazy dream I have the following errors "root@xxx:/workspace/stable-diffusion-webui# fuser -...

Furkan Gözükara SECourses•10/24/23, 11:00 PM

you can ignore Permission denied it is not important

Furkan Gözükara SECourses•10/24/23, 11:00 PM

it killed web ui : 3000/tcp: 91

CCrazy dream Do somebody would know any forum/discord for very newbie please ?

Furkan Gözükara SECourses•10/24/23, 11:00 PM

what help you need?

FFurkan Gözükara SECourses it killed web ui : 3000/tcp: 91

Crazy dream•10/24/23, 11:03 PM

I tried a fourth time to follow the tutoriel but I think my problem from an incompatibility between torchvision and xFormers. But I don't know how to solve it.. I tried the latest version (as show in the video) but it didn't work. I tried the specific pip install xformers==0.0.20 on git but it didn't work either...

CCrazy dream I tried a fourth time to follow the tutoriel but I think my problem from an inco...

Furkan Gözükara SECourses•10/24/23, 11:04 PM

that extension is outdated

Furkan Gözükara SECourses•10/24/23, 11:04 PM

so you need to use exact versions

Furkan Gözükara SECourses•10/24/23, 11:04 PM

written on github

Furkan Gözükara SECourses•10/24/23, 11:04 PM

auto installer on patreon doing that automatically for you

FFurkan Gözükara SECourses what help you need?

Crazy dream•10/24/23, 11:06 PM

General and free help. I know discord is more uptodate but I would prefer a kind of forum where we can follow a topic without losting ourself in messages on different subjects

CCrazy dream General and free help. I know discord is more uptodate but I would prefer a kind...

Furkan Gözükara SECourses•10/24/23, 11:07 PM

this stuff gets updated like once every week

Furkan Gözükara SECourses•10/24/23, 11:08 PM

look reddit stable diffusion

FFurkan Gözükara SECourses auto installer on patreon doing that automatically for you

Crazy dream•10/24/23, 11:08 PM

Thank you but I want to understand how it works. The exact version are about this : "

Automatic1111 : 68f336bd994bed5442ad95bad6b6ad5564a5409a

Dreambooth ext : c2a5617c587b812b5a408143ddfb18fc49234edf

get latest cuda files list on runpod

apt update

apt list -a libcudnn8-dev" ?

CCrazy dream Thank you but I want to understand how it works. The exact version are about thi...

Crazy dream•10/24/23, 11:10 PM

When should I use " apt update

apt list -a libcudnn8-dev" ?

Zono50•10/24/23, 11:11 PM

Kohya broken

soniahere•10/24/23, 11:15 PM

Hi everyone, I am fascinated by the rerender method of creating vid2vid. Wanted to see if there is any reliable method like this for img2vid. I want to animate the pics I create with ai. I know animatediff and other methods but they all lack consistency and have a lot of flickering

_Captain_•10/24/23, 11:17 PM

@Dr. Furkan Gözükara just tried your sdxl dreambooth workflow with textencoder training. Do I understand it right, that you set the general learning rate to 0.000002 and override the unet learning rate afterwards with --block_lr 1e-5,... so that the texencoder lr uses the 0.000002 and the unet 1e-5?

CCrazy dream Thank you but I want to understand how it works. The exact version are about thi...

Furkan Gözükara SECourses•10/24/23, 11:20 PM

do gitcheckout to those versions

Furkan Gözükara SECourses•10/24/23, 11:20 PM

delete venv of auto1111 to rebuild