Software Engineering Courses (SECourses)•2y ago

Prepping a dataset right now of a human subject and most of the quality photos I have are group shot

Prepping a dataset right now of a human subject and most of the quality photos I have are group shots. Is there a way I can modify/psd the group shots to work for training purposes?

Superman•7/15/24, 7:20 AM

Can someone please tell what base model should i use to train a face like this (half cartoon 3d render and half human like)

I was made using faceswap with an cartoon face

SSuperman Can someone please tell what base model should i use to train a face like this (...

Maxivi•7/15/24, 8:20 AM

https://civitai.com/models/216439/artium

Artium - v2.0 | Stable Diffusion Checkpoint | Civitai

A R T I U M An offshoot from my OMNIUM model, geared more towards an illustrative and painterly style, still very much infused with Omnium's DNA in...

TTheToday If I already have a trained model but I want to add more poses/expressions, woul...

Furkan Gözükara SECourses•7/15/24, 8:35 AM

i would improve dataset and do training again

Yyoziv Prepping a dataset right now of a human subject and most of the quality photos I...

Furkan Gözükara SECourses•7/15/24, 8:36 AM

i would avoid

Furkan Gözükara SECourses•7/15/24, 8:36 AM

because you will get mixed faces

SSuperman Can someone please tell what base model should i use to train a face like this (...

Furkan Gözükara SECourses•7/15/24, 8:36 AM

i would download best models from civitiai

Furkan Gözükara SECourses•7/15/24, 8:36 AM

and test my prompt

Furkan Gözükara SECourses•7/15/24, 8:36 AM

and see which one yields best ones

Furkan Gözükara SECourses•7/15/24, 8:36 AM

for generic prompts

J2_Nedo•7/15/24, 10:08 AM

Question, I have had very good results with Onetrainer SD1.5 finetuning and LoRa extraction, now I am trying it with SDXL models, finetuning works very well,
but when I extract LoRa the similarity of the person is much weaker, this is not the case with the SD1.5 extracted LoRAs, do I just have to train longer here ?

daroth•7/15/24, 10:17 AM

btw, how big of a quality difference are we talking about when it comes to direct lora draining vs full checkpoint training + extraction?

Ddaroth btw, how big of a quality difference are we talking about when it comes to direc...

J2_Nedo•7/15/24, 10:28 AM

at least with my datasets i get the best results with full checkpoint training + extraction. How big the difference? i can see it, if extraction is 10 then i would give a 6 or 7

Fireart•7/15/24, 11:36 AM

Hello! I would like to ask for some advice. How can I achieve the best settings for training SDXL Lora on OneTrainer, and what should I pay attention to? The training set consists of approximately 80-150 images of artist’s works with inconsistent aspect ratios. Thank you!

MMaxivi hopefully massed compute gets the 6000ada soon for even faster speeds

NicB•7/15/24, 2:28 PM

The L40 & L40S are very similar to the 6000 ADA. We have a few of those available right now.

The L40 & L40S have a better server CPU and Ram as well.

NNicB The L40 & L40S are very similar to the 6000 ADA. We have a few of those availabl...

Maxivi•7/15/24, 4:07 PM

i think in terms of pricing with runpod both the l40 and 6000ada are quite cheap 79cents per hour

but on massed compute its 1.15 for the l40

MMaxivi i think in terms of pricing with runpod both the l40 and 6000ada are quite cheap...

NicB•7/15/24, 9:25 PM

Apples to apples would be secure cloud pricing on runpod.

JJ2_Nedo Question, I have had very good results with Onetrainer SD1.5 finetuning and LoRa...

Furkan Gözükara SECourses•7/15/24, 10:19 PM

it shouldnt be case

Furkan Gözükara SECourses•7/15/24, 10:19 PM

last time i tested 128 rank lora was very close

Ddaroth btw, how big of a quality difference are we talking about when it comes to direc...

Furkan Gözükara SECourses•7/15/24, 10:20 PM

i think a lot

Furkan Gözükara SECourses•7/15/24, 10:20 PM

extraction was way better for me

FFireart Hello! I would like to ask for some advice. How can I achieve the best settings ...

Furkan Gözükara SECourses•7/15/24, 10:20 PM

i dont have best hyper configuration for lora training

Furkan Gözükara SECourses•7/15/24, 10:20 PM

it requires some research

Furkan Gözükara SECourses•7/15/24, 10:20 PM

you need to find best learning rate and other parameters

Furkan Gözükara SECourses•7/15/24, 10:20 PM

but you can use our fine tuning parameters

Furkan Gözükara SECourses•7/15/24, 10:20 PM

and extract lora after training if necessary

FFurkan Gözükara SECourses last time i tested 128 rank lora was very close

J2_Nedo•7/16/24, 7:35 AM

thank for the info, i will try the same dataset training on different SDXL Models to see if there is a difference.

JJ2_Nedo Question, I have had very good results with Onetrainer SD1.5 finetuning and LoRa...

abtx•7/16/24, 8:22 PM

Make the threshold lower. For me 0.0001 at most.

Aabtx Make the threshold lower. For me 0.0001 at most.

Furkan Gözükara SECourses•7/16/24, 9:29 PM

true make sure that diff limit low

Furkan Gözükara SECourses•7/16/24, 9:29 PM

like 0.000001

J2_Nedo•7/17/24, 8:25 AM

ok i trained on the standard SDXL Model now, Epoch around 150 Model is really good, but the LoRA export still not close to the finetuning, i tested different threshold settings, 0.001 - 0.000001 it didn't helped. i will try later different ranks (128), do i need to change both rank and alpha to the same value?

JJ2_Nedo ok i trained on the standard SDXL Model now, Epoch around 150 Model is really go...

Maxivi•7/17/24, 9:37 AM

are you training the text encoder as well? --train_text_encoder --learning_rate_te2 0 ??

JJ2_Nedo ok i trained on the standard SDXL Model now, Epoch around 150 Model is really go...

Furkan Gözükara SECourses•7/17/24, 9:54 AM

i think give it a try

Furkan Gözükara SECourses•7/17/24, 9:54 AM

i cant test right now but i plan research and a video for this

MMaxivi are you training the text encoder as well? --train_text_encoder --learning_rate_...

J2_Nedo•7/17/24, 10:20 AM

only Text Encoder 1

Maxivi•7/17/24, 10:49 AM

try an older version of kohya

Maxivi•7/17/24, 10:49 AM

like 22.6.2

MMaxivi like 22.6.2

J2_Nedo•7/17/24, 11:07 AM

uh ok, i first need to learn how to rollback, never tried before

Maxivi•7/17/24, 11:07 AM

git checkout 5d77bf409ff6a238aa2da94cf7910f97b8f1f1e9

Maxivi•7/17/24, 11:08 AM

or download the zip and make a extra copy of kohya

Maxivi•7/17/24, 11:08 AM

kohya_ss-22.6.2.zip7.47MB

J2_Nedo•7/17/24, 11:08 AM

ok thax for the help, will try when at home later tonight!

JJ2_Nedo ok thax for the help, will try when at home later tonight!

Furkan Gözükara SECourses•7/17/24, 11:15 AM

i have shown in this video how to checkout : https://youtu.be/-NjNy7afOQ0

YouTubeSECourses

Essential AI Tools and Libraries: A Guide to Python, Git, C++ Compi...

Welcome to the Pillars of AI Application Libraries: How to Install Tutorial!

Are you ready to embark on a journey of installing AI libraries and applications with ease?

In this video, we'll guide you through the process of installing Python, Git, Visual Studio C++ Compile tools, and FFmpeg on your Windows 10 machine.

We'll also show you ho...

Furkan Gözükara SECourses•7/17/24, 11:15 AM

18:18 How to do a git checkout to switch different branches such as dev branch

Furkan Gözükara SECourses•7/17/24, 11:15 AM

the lo

Furkan Gözükara SECourses•7/17/24, 11:15 AM

logic is same for versions

Bill Meeks•7/17/24, 11:14 PM

Create an original character LoRA in one shot following this tutorial. https://everlyheights.tv/stablediffusion/create-an-original-character-lora-with-everly-heights-character-maker-free-template/

Sergio A.•7/19/24, 4:39 AM

I'm seeking an expert in Generative AI to create a text-to-image generation model prototype for me. Here are the key outcomes I'm expecting:

The AI model must create cartoon-style images. A specialist with experience in this area will be a great fit. Knowledge in digital art or animation is a plus.
The model should be fine tunes with images of mascots of different brands like McDonalds, Amul etc and final output should of genAI generated different images of the mascots
The images should be small in size, less than 500x500 pixels. Experience with condensed data optimization and high-quality rendering is desired.
I want the generated images to be in color. Proficiency in the standard RGB color model will be essential to meet this requirement.

I look forward to seeing your creative application of Artificial Intelligence to meet these specifications.

SSergio A.I'm seeking an expert in Generative AI to create a text-to-image generation mode...

Furkan Gözükara SECourses•7/19/24, 9:14 AM

i sent dm to you

MMaxivi try an older version of kohya

J2_Nedo•7/19/24, 9:28 AM

hello, i tested with Kohya_ss 22.6.2 and the first results are promising. But i will do more tests over the weekend with different Models. Thanks for the Info!

JJ2_Nedo hello, i tested with Kohya_ss 22.6.2 and the first results are promising. But i ...

Maxivi•7/19/24, 4:55 PM

glad to know, i think their is an issues with the new kohya because they intrduced many news options like huber schedule and other settings that the lora is not extracting properly