Software Engineering Courses (SECourses)•14mo ago

Yes sirr 🫡 . Had to override the default =0.

Yes sirr

. Had to override the default =0.

SShamsi Yes sirr 🫡 . Had to override the default =0.

Furkan Gözükara SECourses•11/19/24, 12:20 AM

yep

FFurkan Gözükara SECourses yes you need to enter max steps count

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/19/24, 6:54 AM

Perfect, just what I was gonna ask too

Got my answer quickly here. Thanks

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/19/24, 10:21 AM

Here's a challenge I want to complete. I have a fine-tuned checkpoint for my character, but I want my characters body to be taken from another character. So I trained a lora using pics of the other character, but the lora also includes that characters face. What do you guys suggest I do to get best result? I don't want my face changed because of this lora. So is using the <segment...> syntax in swarmui a way to go, or should I change the training images such that the head is cut off?

Appreciate this community so much, I learn something every day!

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭Here's a challenge I want to complete. I have a fine-tuned checkpoint for my cha...

Furkan Gözükara SECourses•11/19/24, 10:45 AM

i dont think it will work

Furkan Gözükara SECourses•11/19/24, 10:45 AM

you can inpaint face of another character

Rockerx•11/19/24, 11:46 AM

@Furkan Gözükara SECourses reporting 2 things.

now when I install v44 or v14 scripts it does update kohya and I am getting correct step count.
in your lora configs you are using additional parameters for block swap. this throws an exit 0 error and training exits. the UI in lora in latest kohya now has block swap option in UI. if I manually set it to what your config has and remove the additional parameter the training works but not sure if as the UI option says (use with fused back pass) which is not available in UI. but the training works with block swap if I set in UI and remove your addition parameters setting.
Can you fix the configs so that it works for lora ? have not tested the finetune in new config but see if that also needs update etc.

wardensc2•11/20/24, 1:57 AM

hi @Furkan Gözükara SECourses I just download dreambooth v15 few days ago now you already uploaded v16. Is there something new in this new version ?

wardensc2•11/20/24, 2:05 AM

And I also have a question about training full dreambooth checkpoint @Furkan Gözükara SECourses . Since you said training person with dreambooth should separate into 2 part of training. Part 1 to take the epoch which is flexible to style. Part 2 is train new epoch with the base is flexible style model from part 1.
What is the different with continue training compare with separte with 2 parts like this. since we used constant learning rate.
For example: Part 1 - I choose epoch 30 to use as based model to retrain. Part 2 - I train 100 more epoch with based model in Part 1. But that just similar to train 130 epoch from Part 1 right ? Because we use constant learning rate.

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/20/24, 8:54 AM

So as a general rule, the more training images you have of 1 character, the less epochs you would need to get good result. Am I getting the logic right? So 300 pics you might get good result with 10 or 20 epochs, but with 15 pics, you will need more epochs, like 100-200.

Wwardensc2 And I also have a question about training full dreambooth checkpoint <@205854764...

Furkan Gözükara SECourses•11/20/24, 10:11 AM

yes both should be same continue

Wwardensc2 hi <@205854764540362752> I just download dreambooth v15 few days ago now you alr...

Furkan Gözükara SECourses•11/20/24, 10:12 AM

no changes for fine tuning only dowgnrade torch to 2.5.0

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭So as a general rule, the more training images you have of 1 character, the less...

Furkan Gözükara SECourses•11/20/24, 10:12 AM

true

dxqb•11/20/24, 10:34 AM

@Furkan Gözükara SECourses is the OneTrainer UI issue still existing? Is there an issue on github?

Ddxqb <@205854764540362752> is the OneTrainer UI issue still existing? Is there an iss...

Furkan Gözükara SECourses•11/20/24, 10:52 AM

last time i tested it was but i didnt test recentyly

Furkan Gözükara SECourses•11/20/24, 10:52 AM

planning to test after we update our massed compute image

dxqb•11/20/24, 11:02 AM

if you open an issue I can have a look, but please provide as much info as you can so I can replicate it without massed compute

mkafil2001•11/20/24, 1:38 PM

Do any have experience in training yourself with dreambooth sdxl model ....
This the params i use

but my in the image generation i get completly different image

Whether using vae model is the mistake hera ....
DR.furkan gozukara u only can help me out

I was stuck in it for past 1 month

I need to train myself and reproduce it like professional image

dxqb•11/20/24, 4:46 PM

@Furkan Gözükara SECourses are you in contact with anyone who is doing large finetunes like RealVis/Juggernaut, and is attempting to do the same for Flux? I might have something that helps.

Xcorpion•11/20/24, 6:29 PM

@Furkan Gözükara SECourses ss_network_args": "{"loraplus_lr_ratio": "16", "loraplus_text_encoder_lr_ratio": "4", "dropout": null}"
do you know what is this? if know how to do this?

Mmkafil2001 Do any have experience in training yourself with dreambooth sdxl model .... Thi...

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

looks to me very random parameters

Ddxqb <@205854764540362752> are you in contact with anyone who is doing large finetune...

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

not any sadly

XXcorpion <@205854764540362752> ss_network_args": "{\"loraplus_lr_ratio\": \"16\", \"lorap...

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

i dont know

ShamsiOP•11/20/24, 11:54 PM

@Furkan Gözükara SECourses Doc, When preparing the images, is the resizing step necessary? I cropped using your tool and I'm not training on faces. Most images to train on are larger than 1024p and in various aspect ratios. in that case, should I change the max resolution setting in Kohya?

SShamsi <@205854764540362752> Doc, When preparing the images, is the resizing step neces...

Furkan Gözükara SECourses•11/21/24, 12:16 AM

if you dont resize

Furkan Gözükara SECourses•11/21/24, 12:16 AM

they will be random aspect ratios

Furkan Gözükara SECourses•11/21/24, 12:16 AM

depending on subject cropping

FFurkan Gözükara SECourses they will be random aspect ratios

ShamsiOP•11/21/24, 12:20 AM

In kohya, do I change the max resolution or maybe enable buckets? images are around 1000*1900 few are large and some are smaller

SShamsi In kohya, do I change the max resolution or maybe enable buckets? images are aro...

Furkan Gözükara SECourses•11/21/24, 12:22 AM

i would crop myself to all 1024x1024 it works best

Furkan Gözükara SECourses•11/21/24, 12:22 AM

what you are training?

ShamsiOP•11/21/24, 12:23 AM

clothes types. traditional dresses

FFurkan Gözükara SECourses i would crop myself to all 1024x1024 it works best

ShamsiOP•11/21/24, 12:23 AM

not really an option

SShamsi clothes types. traditional dresses

Furkan Gözükara SECourses•11/21/24, 12:24 AM

ok in that case

Furkan Gözükara SECourses•11/21/24, 12:24 AM

make it like 2048x2048 and enable bucket

Furkan Gözükara SECourses•11/21/24, 12:25 AM

but i cant say how good it will work

Furkan Gözükara SECourses•11/21/24, 12:25 AM

a bit experimental

Furkan Gözükara SECourses•11/21/24, 12:25 AM

i trained a style at 1536px and it didnt learn style accurately

ShamsiOP•11/21/24, 12:25 AM

haha, will let you know the results

ShamsiOP•11/21/24, 12:26 AM

Ty doc and gn

FFurkan Gözükara SECourses looks to me very random parameters

mkafil2001•11/21/24, 1:04 AM

How can i select the better one

Mmkafil2001 How can i select the better one

Furkan Gözükara SECourses•11/21/24, 1:04 AM

i have sdxl dreambooth good config

Furkan Gözükara SECourses•11/21/24, 1:04 AM

they are shown in this tutorial

Furkan Gözükara SECourses•11/21/24, 1:04 AM

https://youtu.be/0t5l6CP9eBg

YouTubeSECourses

Full Stable Diffusion SD & XL Fine Tuning Tutorial With OneTrainer ...

In this tutorial, I am going to show you how to install OneTrainer from scratch on your computer and do a Stable Diffusion SDXL (Full Fine-Tuning 10.3 GB VRAM) and SD 1.5 (Full Fine-Tuning 7GB VRAM) based models training on your computer and also do the same training on a very cheap cloud machine from MassedCompute if you don't have such compute...