Software Engineering Courses (SECourses)•10mo ago

Hi Dr Furkan, Shaun here. I'm planning to fine-tune an SDXL model checkpoint on a large number of im

Hi Dr Furkan, Shaun here. I'm planning to fine-tune an SDXL model checkpoint on a large number of images, hoping to improve its base performance. Some examples are like JuggernautXL , RealVis XL etc. Were they trained using dreambooth as well? Is dreambooth the best method to train a general model? As a lot of the examples of dreambooth seem to be only about training on a specific subject.

Ssai_150 I'm testing the model results, I see in some generated photos the background and...

Furkan Gözükara SECourses•3/16/25, 12:15 PM

nope

Furkan Gözükara SECourses•3/16/25, 12:16 PM

in inference instead change your prompt

Furkan Gözükara SECourses•3/16/25, 12:16 PM

define them

Furkan Gözükara SECourses•3/16/25, 12:16 PM

so they wont come

Furkan Gözükara SECourses•3/16/25, 12:16 PM

define something else instead like new background new clothing

Furkan Gözükara SECourses•3/16/25, 12:16 PM

actually if you improve dataset that is best

Ssunflowerm Hi Dr Furkan, Shaun here. I'm planning to fine-tune an SDXL model checkpoint on ...

Furkan Gözükara SECourses•3/16/25, 12:16 PM

yes they do dreambooth without regularization thus it is fine tuning

Furkan Gözükara SECourses•3/16/25, 12:17 PM

if you prepare a large enough dataset with proper captions it should work

Sspirit18 Huge Thanks to <@205854764540362752> 🥹

daquan0435•3/16/25, 2:14 PM

What ıs the promt?

Jackie•3/18/25, 2:32 AM

Hello, I am new to training DreamBooth. May I ask if the Kaggle Notebook on Patreon can be used to train the Illustrious XL model?

FFurkan Gözükara SECourses actually if you improve dataset that is best

sai_150•3/18/25, 3:09 AM

I'm using 13 images now with 4 unique backgrounds and clothes, can you tell me how many images are needed? how many steps should I train? config file has train batch size as 2, should I try train batch size =1. Does it help if I remove the background in the train images?

gio3000.•3/18/25, 7:58 AM

On massive compute, is it possible for dreamboth to access my dataset in a Cloud? Space is too limited on massive compute

JJackie Hello, I am new to training DreamBooth. May I ask if the Kaggle Notebook on Patr...

Furkan Gözükara SECourses•3/18/25, 1:32 PM

yes but it may require new hyper parameters

Furkan Gözükara SECourses•3/18/25, 1:32 PM

kaggle book we have supports SDXL with dual GPU

Ssai_150 I'm using 13 images now with 4 unique backgrounds and clothes, can you tell me h...

Furkan Gözükara SECourses•3/18/25, 1:32 PM

more better

Furkan Gözükara SECourses•3/18/25, 1:32 PM

13 low

Ggio3000.On massive compute, is it possible for dreamboth to access my dataset in a Cloud...

Furkan Gözükara SECourses•3/18/25, 1:32 PM

yes

Furkan Gözükara SECourses•3/18/25, 1:32 PM

i recommend hugging face but you can use any

Furkan Gözükara SECourses•3/18/25, 1:32 PM

by the way first you have to download

Furkan Gözükara SECourses•3/18/25, 1:32 PM

to massed compute

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•3/19/25, 6:26 AM

Hey. Is it now best to use sdpa instead of xformers for both lora and fine tuning? Doing both flux and sdxl

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•3/19/25, 6:26 AM

@Furkan Gözükara SECourses

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭Hey. Is it now best to use sdpa instead of xformers for both lora and fine tunin...

Furkan Gözükara SECourses•3/19/25, 8:58 AM

i feel almost same but xformers just a little bit better

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•3/19/25, 2:03 PM

Thanks Dr

ProRost•3/19/25, 4:04 PM

I have two questions about DreamBooth training:

Is it possible to install Super (super resolution upscaler) on MacBook? As far as I understand, Super can be run on macOS, but it is important to note that some features may require CUDA support (which is limited on MacBook). If you have already used it on Mac, please share your experience!
Is it possible to do a full SDXL Dreamboot or Dreamboot FLUX training via Google Colab? Or, in your experience, is it better to use cloud services such as Massed Compute or RunePod (I heard recommendations from Doc)?

I would be grateful for advice on the best option.

Thanks!

JP.Lonb•3/19/25, 5:36 PM

Hello Furkan,
Here is JP. I recently contacted you on YouTube to report a difficulty that I have performing a with Dreambooth Flux training with Kohya and you suggested to contact you through discord, thus here I am. Let me just first say that I really appreciate what you are doing here for all of us noobs interested in AI creation. I am a 60 years old doctoral program manager (with also several PhD students coming from Turkey!) in chemistry and I only discovered recently this interesting area of AI picture creation. How fascinating! But without your great tutorials, I would certainly not manage! I watch them again and again until I finally manage to solve my problems.

JP.Lonb•3/19/25, 5:36 PM

Coming to the latest problem, I do not manage to create any Flux dreambooth with your Kohya installation on my PC, unfortunately. I was already successful with Kohya Flux Lora trainings, also with OT Finetuning SDXL trainings, all loca trainings on my PC, but now there seem to be a problem (crash) at the moment Kohya wants to save the first safesensors file. I am running the training on a RTX4080 super with 16GB VRAM (could not afford more, unfortunately), choose your 16GB config (which leads to a VRAM usage of about 14,4 GB, shared usage of 18,3/31.5). Speed is approximately 5s/it, which is ok, I think. I tried with different number of steps (200, 100 ….down to much less), had started with 124 pics (1 repeat), am now down to 15 to for testing purpose. Everything goes fine until Kohya tries to save the first intermediate epoch/safetensor file. Then it crashes, terminates the training, as if it could not save the file. I tried to have it saved at many different places, it makes no difference. Could it be that the size is too big? As already mentioned, I had no problem with the Kohya Flux Loras and OT SDXL Finetunings.

JP.Lonb•3/19/25, 5:37 PM

When Kohya tries to save the first safetensors file, it does not say anything, but just crashes. My screen goes dark for a couple of minute. Then I can again check what happened. The console says that the training has ended. I can end you the full log, if you want to try to see what happened. It starts with “--- Logging error ---
Traceback (most recent call last):
File "C:\Python310\lib\logging__init.py", line 1104, in emit
self.flush()
File "C:\Python310\lib\logging__init.py", line 1084, in flush
self.stream.flush()
OSError: [Errno 22] Invalid argument
Call stack:
File "C:\Python310\lib\threading.py", line 973, in _bootstrap
self._bootstrap_inner()
etc. “

JP.Lonb•3/19/25, 5:37 PM

Following the advices of ChatGPT, I check for critical errors in the “Windows Event Viewer Logs“ and found indeed several ones at the time of the crash. I can send you the evtx file, if it helps. I also looked into the “reliability monitor” and there you can also see warnings and application failures at the crash time. What can be the reason?!
Thanks for your help in advance!
Greetings from Berlin
JP

PProRost I have two questions about DreamBooth training: 1. Is it possible to install Sup...

Furkan Gözükara SECourses•3/19/25, 6:54 PM

you mean SUPIR i suppose. our app works on windows but i have no idea can be run on mac which i dont think so

Furkan Gözükara SECourses•3/19/25, 6:55 PM

but you can use on even a free kaggle

Furkan Gözükara SECourses•3/19/25, 6:55 PM

you can do SDXL on kaggle for free

Furkan Gözükara SECourses•3/19/25, 6:55 PM

on colab look kaggle code

Furkan Gözükara SECourses•3/19/25, 6:55 PM

and you can run on paid colab

JJP.Lonb Coming to the latest problem, I do not manage to create any Flux dreambooth with...

Furkan Gözükara SECourses•3/19/25, 6:56 PM

hi i would recommend this

Furkan Gözükara SECourses•3/19/25, 6:56 PM

uninstall your python and reinstall as shown in this video

Furkan Gözükara SECourses•3/19/25, 6:56 PM

and reinstall kohya too

Furkan Gözükara SECourses•3/19/25, 6:56 PM

are you using our kohya installer,?

Furkan Gözükara SECourses•3/19/25, 6:56 PM

https://youtu.be/DrhUHnYfwC0

YouTubeSECourses

How to Install Python, CUDA, cuDNN, C++ Build Tools, FFMPEG & Git T...

If you are interested in using AI, generative AI applications, open source applications in your computer, than this is the most fundamental and important tutorial that you need. In this tutorial I show and explain how to properly install appropriate Python versions accurately, how to switch between different Python versions, how to install diffe...

ProRost•3/20/25, 11:29 AM

and tell me where to find kaggle code? my main task is to make a photo of the generation with the face of a real person, but replace the body with another. Is there an option how to do this? gpt chat tells me that I should do a full training of dreambooth, then somehow place it in 3D, longer modify or replace the body. What do you advise? Also, I will need generations without censorship (18+)

PProRost and tell me where to find kaggle code? my main task is to make a photo of the ge...

Furkan Gözükara SECourses•3/20/25, 11:30 AM

forget kaggle for anything NSFW and i dont support NSFW either

Furkan Gözükara SECourses•3/20/25, 11:30 AM

kaggle will ban your account immediately

ProRost•3/20/25, 12:55 PM

I agree Doc, I am also against it but I was asked to help with this request. If without NSFW, what would be the best way?

PProRost I agree Doc, I am also against it but I was asked to help with this request. If ...

Furkan Gözükara SECourses•3/20/25, 2:01 PM

kaggle works for SDXL and SD 1.5 training

ProRost•3/20/25, 2:25 PM

but based on this video https://www.youtube.com/watch?v=FvpWy1x5etM I concluded that Flax Fal Fine tuning has much better detailing. If you compare SDXL and SD 1.5, the XL version will win in detailing by a lot. Honestly, I am very grateful for your work and contribution to the development of neural networks.

YouTubeSECourses

FLUX Full Fine-Tuning / DreamBooth Training Master Tutorial for Win...

If you want to train FLUX with maximum possible quality, this is the tutorial looking for. In this comprehensive tutorial, you will learn how to install Kohya GUI and use it to fully Fine-Tune / DreamBooth FLUX model. After that how to use SwarmUI to compare generated checkpoints / models and find the very best one to generate most amazing image...

PProRost but based on this video https://www.youtube.com/watch?v=FvpWy1x5etM I concluded ...

Furkan Gözükara SECourses•3/20/25, 2:45 PM

FLUX is 10 times better than SDXL

gio3000.•3/20/25, 3:35 PM

Hi everyone and @Furkan Gözükara SECourses
I have a paired dataset with images before and after retouching (color correction, defect removal on face and body, skin color amelioration). I want to train a generative model to automate this process at scale.

Would Pix2Pix (GANs) be the best choice, or are diffusion models, flux , vision transformers, or other supervised methods better for high-fidelity, production-ready retouching?

Looking for insights on the most effective approach. Thanks!

Ggio3000.Hi everyone and <@205854764540362752> I have a paired dataset with images befor...

Furkan Gözükara SECourses•3/20/25, 6:31 PM

sadly i dont know what can be best way for automating atm

umrfarooq•3/21/25, 8:55 PM

Hi @Furkan Gözükara SECourses ,

I need some advice as I have been struggling with Lora training for a product. I tried with and without captions (without was worse).

Problems I'm Facing

Trigger words don't work consistently. For example, if I use "KOKUMKF2512NVYP2," sometimes I get some kind of device (see trigger_word_image) and sometimes a dress but with different colors and the same pattern.
When I use prompts where I explain the dress and background, I get a blue dress, but each time the dress color appears in different shades.
Hands are distorted a lot.
When I prompt it to generate on different body types, it doesn't work.

## What I Have Tried So Far

No Caption Lora:

batch_size: 2
steps: 4000

batch_size: 2
steps: 4000

Captioned Lora (see screenshot for dataset):

batch_size: 2
steps: 3500
sample caption: a person seated on a wooden chair against a dark gray paneled wall. The background is simple with wooden flooring visible. The model is wearing KOKUMKF2512NVYP2 with a KOKUMKF2512NVYP2DUPATTA draped across their lap and extending to the floor. The model is accessorized with pearl-like earrings, bracelets, and embellished footwear. Their hands are positioned casually, with one resting on their lap and the other on the arm of the wooden chair. The setting appears to be an indoor studio environment with minimalist decor, highlighting the seated figure against the contrasting background.

batch_size: 2
steps: 3500
sample caption: a person seated on a wooden chair against a dark gray paneled wall. The background is simple with wooden flooring visible. The model is wearing KOKUMKF2512NVYP2 with a KOKUMKF2512NVYP2DUPATTA draped across their lap and extending to the floor. The model is accessorized with pearl-like earrings, bracelets, and embellished footwear. Their hands are positioned casually, with one resting on their lap and the other on the arm of the wooden chair. The setting appears to be an indoor studio environment with minimalist decor, highlighting the seated figure against the contrasting background.

Mixed Lora of cloth and model head (see screenshot for dataset):

batch_size: 1
steps: 4000
sample caption for model headshot image: The image shows a zoya_aurage looking off to the side with a neutral expression. The background is a solid gray color creating a minimal, studio-like setting. zoya_aurage is positioned in a profile-like pose with bare shoulders visible, creating a clean and professional portrait composition.

batch_size: 1
steps: 4000
sample caption for model headshot image: The image shows a zoya_aurage looking off to the side with a neutral expression. The background is a solid gray color creating a minimal, studio-like setting. zoya_aurage is positioned in a profile-like pose with bare shoulders visible, creating a clean and professional portrait composition.

trigger_word_image92.12KB

trigger_word_image_256.77KB

prompt_sample663B

umrfarooq•3/21/25, 9:06 PM

Thank you so much for doing amazing research and sharing with us. And Sorry for the little long message I just want to share as much detail as possible. And is full finetuning is a good choice for this specific use case or not?

Here is some more details:

For the dataset:
Dress: 16 images
Model/Person: 9 headshot images generated by comfy consistent character workflow

I also tried using both lora (dress and person) to create images in SwarmUI but results are disappointing. The major importance for me is to get the pattern of dress, color of dress and person wearing wearing that dress. Please let me know If you need any other details. Please guide me to the right direction. Thanks again

Hi Dr Furkan, Shaun here. I'm planning to fine-tune an SDXL model checkpoint on a large number of im

Problems I'm Facing

No Caption Lora:

Captioned Lora (see screenshot for dataset):

Mixed Lora of cloth and model head (see screenshot for dataset):

Similar Threads

Problems I'm Facing

No Caption Lora:

Captioned Lora (see screenshot for dataset):

Mixed Lora of cloth and model head (see screenshot for dataset):