Software Engineering Courses (SECourses)•2y ago

I used a Learning rate of .001, thew results were much better! .001 might be high but because I'm do

I used a Learning rate of .001, thew results were much better! .001 might be high but because I'm doing parody I don't really need exact I just need to capture their likeness and make them something else

Enlil Wolf•12/27/23, 3:23 PM

@Dr. Furkan Gözükara - Dr, when you get the chance, please check the DM

LLorelai AI I used a Learning rate of .001, thew results were much better! .001 might be hig...

Furkan Gözükara SECourses•12/27/23, 3:24 PM

ye it changes according to settings

B0nd•12/27/23, 4:35 PM

When training for SDXL DreamBooth or Lora in KoyhaSS what is generally the best way, for likeness and flexibility?

With or without regularization images?
With or without captioning the dataset?

FFurkan Gözükara SECourses you need to use our 16gb config

DigitalArt_Lover•12/27/23, 5:41 PM

I need to know, the 16gb config, can i get it from you by any chance?? ..i would buy the Patreon way long ago, but for my country i could pay, because of the payment method restrictions.

BB0nd When training for SDXL DreamBooth or Lora in KoyhaSS what is generally the best ...

Furkan Gözükara SECourses•12/27/23, 7:54 PM

it depends purpose. i made a good comparison here recently

Furkan Gözükara SECourses•12/27/23, 7:55 PM

https://civitai.com/articles/3510/dreambooth-training-on-sdxl-anime-model-and-comparing-ground-truth-regularization-images-effect

DreamBooth Training on SDXL Anime Model & Comparing Ground Truth Re...

It was requested of me to test ground truth Regularization / Classification images effect during Stable Diffusion XL (SDXL) DreamBooth training. So...

DDigitalArt_Lover I need to know, the 16gb config, can i get it from you by any chance?? ..i would...

Furkan Gözükara SECourses•12/27/23, 7:55 PM

hello. sure i accept crypto and other methods too

Furkan Gözükara SECourses•12/27/23, 7:55 PM

you can dm me

Benjamin•12/27/23, 8:14 PM

@Dr. Furkan Gözükara did I get it right, you are now training with 150 repeats, 1 epoch for 13 training images and save steps instead of epochs?

Benjamin•12/27/23, 8:15 PM

so 150 * 13 * 1 *2 = 3900 steps?

Digital [Starburst]•12/27/23, 8:21 PM

I loaded the styles.csv into Automatic and have gotten some very interesting results using the updated Juggernaut 7 model. The advantage to this model is the refiner is built into the model, so you don’t need a separate refiner. The disadvantage to using the styles in Automatic is that you cannot combine styles in one image. If you select multiple styles you will get a batch for each style you choose.

Digital [Starburst]•12/27/23, 8:24 PM

In the Fooocus UI you can combine styles for more variety in your art. You can also use the same model by manually adding it. I personally like to select one style and generate a prompt to see the results before adding another style or switching styles. The Fooocus Masterpiece style is my favorite so far.

Digital [Starburst]•12/27/23, 8:26 PM

Combining two styles is, in essence, a completely different style. As you add styles to a prompt you get unexpected results.

Digital [Starburst]•12/27/23, 8:28 PM

The caveat to combining styles is one may completely cancel out others.

Digital [Starburst]•12/27/23, 8:33 PM

I take my images, load them into Corel Painter or Corel Paintshop Pro and adjust them, make them fit an 8.5 X 11 sheet of paper or 11 X 17 sheet for printing. I also may completely alter the image because I don’t want photographs even though I want to study photography in order to write good prompts. At the moment I borrow other people’s prompts and modify them. I don’t want to be reliant on someone else’s prompts forever though.

Digital [Starburst]•12/27/23, 8:35 PM

It seems that the general consensus for most AI developers is they want photographic art. That’s all well and good but they are really limiting themselves in my opinion as SD is capable of so much more.

Digital [Starburst]•12/27/23, 8:38 PM

True. Not without the use of other digital art tools but the results I am seeing can’t be ignored.

Digital [Starburst]•12/27/23, 8:39 PM

Digital [Starburst]•12/27/23, 8:42 PM

That is a raw image straight from Fooocus UI using 3 styles but the prompt itself also specifies a style which overrides apparently the selected styles.

allen•12/27/23, 10:37 PM

Howdy all - I'm trying to get MagicAnimate working in Runpod, can anyone point me to a thread or discussion?

allen•12/27/23, 10:38 PM

I tried it in comfyUI local install and it just seems buggy, the nodes are anyway, and then I realized the good Dr had already put together a better implementation. I have the Gradio space working right now in a runpod

lajotas.•12/27/23, 10:43 PM

@Dr. Furkan Gözükara , when training a SDXL model with 30 training images of a subject for a person's resemblance, after how many epochs do you expect the images to start looking like the person?

BBenjamin @Dr. Furkan Gözükara did I get it right, you are now training with 150 repeats, ...

Furkan Gözükara SECourses•12/28/23, 12:00 AM

yes

DDigital [Starburst]Click to see attachment

Furkan Gözükara SECourses•12/28/23, 12:01 AM

it is true

Furkan Gözükara SECourses•12/28/23, 12:01 AM

i told foocus developer to add

Furkan Gözükara SECourses•12/28/23, 12:01 AM

testing feature

Furkan Gözükara SECourses•12/28/23, 12:01 AM

he ignored

Aallen Howdy all - I'm trying to get MagicAnimate working in Runpod, can anyone point m...

Furkan Gözükara SECourses•12/28/23, 12:01 AM

hello

Furkan Gözükara SECourses•12/28/23, 12:01 AM

i recorded the video

Furkan Gözükara SECourses•12/28/23, 12:01 AM

hopefully will be published tomorrow

Furkan Gözükara SECourses•12/28/23, 12:01 AM

magic animate tutorial very detailed

Llajotas.@Dr. Furkan Gözükara , when training a SDXL model with 30 training images of a s...

Furkan Gözükara SECourses•12/28/23, 12:02 AM

after 20 epochs it should start to resemable

Furkan Gözükara SECourses•12/28/23, 1:10 AM

my kaggle kohya notebook works directly on google colab pro nice

Furkan Gözükara SECourses•12/28/23, 1:10 AM

just made a consultation to colab pro user

andysingal•12/28/23, 1:43 AM

@Dr. Furkan Gözükara Any good tools to fix deformed faces, limbs and hands when creating animations?

andysingal•12/28/23, 3:00 AM

i tried inpaint + Adetailer but did not work

codingmaniac•12/28/23, 3:44 AM

Hello

codingmaniac•12/28/23, 3:45 AM

I need help with React. I can't seem to get my buttons for my calculator to pop up

codingmaniac•12/28/23, 3:45 AM

Its my first React project btw really new at React(like 4 days ago new)

Aandysingal @Dr. Furkan Gözükara Any good tools to fix deformed faces, limbs and hands when ...

Furkan Gözükara SECourses•12/28/23, 9:50 AM

i replied your message on twitter too. that photo need upscale and overal improvement

Furkan Gözükara SECourses•12/28/23, 9:50 AM

to further fixed

Squid•12/28/23, 9:54 AM

Hello. I'm currently trying to create a face lora after watching your video: https://www.youtube.com/watch?v=TpuDOsuKIBo&t=721s. The results are somewhat disappointing, possibly due to dataset issues, but it's definitely improved since I started experimenting on my own after watching the video. However, I'm wondering if cropping a person's face or full body shot from the background and turning it into a white background would be significantly beneficial? I actually feel like it's ruining the train and result looks weird right now

YouTubeSECourses

Generate Studio Quality Realistic Photos By Kohya LoRA Stable Diffu...

#Kohya SS web GUI DreamBooth #LoRA training full tutorial. You don't need technical knowledge to follow this tutorial. In this tutorial I have explained how to generate professional photo studio quality portrait / self images for free with Stable Diffusion training.

Our Discord server

https://bit.ly/SECoursesDiscord

If I have been of assis...

Generate Studio Quality Realistic Photos By Kohya LoRA Stable Diffu...

FFurkan Gözükara SECourses i replied your message on twitter too. that photo need upscale and overal improv...

andysingal•12/28/23, 1:37 PM

I tried adetail for upscaling but no success

B0nd•12/28/23, 1:42 PM

Whats the best way to upscale your dataset for maximum detail?

FFurkan Gözükara SECourses it is true

Digital [Starburst]•12/28/23, 5:50 PM

Any chance a group of us could try and develop code to win $700,000.00 by deciphering 4 ancient documents found near Pompeii? https://www.msn.com/en-us/news/technology/unlocking-ancient-wisdom-ai-decodes-greek-word-from-herculaneum-scrolls/ar-AA1m3ZUO?ocid=msedgntp&pc=EDGEESS&cvid=5b89fbd3e75548e799f8543b42b70568&ei=16

Digital [Starburst]•12/28/23, 5:51 PM

They are called the Herculaneum Scrolls, and since we are so hyped by AI and that is the key to unlocking the text it might be fun to do it even if we lose.

shortdwarf•12/28/23, 7:19 PM

Has anyone tried training a LoRA or Dreambooth using PNGs of a subject with a transparent background (essentially subject cut out and nothing behind)? I've tried with white background and backgrounds of random colors, both successful to remove associating the training with the background and focus more on the subject, however im curious if anyone has already tried this and could speak about the results
EDIT: When using ohwx man within a prompt, the random color background works quite well. However I just tried a simple "(ohwx man:1.3)" prompt and it mostly worked well however I did receive one photo that was just a purple background, so thats worth noting

BB0nd Whats the best way to upscale your dataset for maximum detail?

shortdwarf•12/28/23, 7:33 PM

From what I've seen online it seems .dat upscalers are the best. However I can't personally speak to this since those are not yet supported by A1111. But looking at this, https://openmodeldb.info/models/4x-FaceUpDAT it seems exceptional

SSquid Hello. I'm currently trying to create a face lora after watching your video: htt...

Furkan Gözükara SECourses•12/28/23, 7:48 PM

what is your GPU?

Furkan Gözükara SECourses•12/28/23, 7:48 PM

SD 1.5 is really weak. SDXL makes huge difference

I used a Learning rate of .001, thew results were much better! .001 might be high but because I'm do

Similar Threads

Similar Threads

Similar Threads