Software Engineering Courses (SECourses)•3y ago

What has better results for realism, drambooth or Lora?

FFurkan Gözükara SECourses https://youtu.be/-xEwaQ54DI4

hozaggers•12/15/23, 10:02 AM

I am using kaggle and I notice the default path to the model is stabilityai/stable-diffusion-xl-base-1.0 - I don't really know where this stabilityai folder is... if I use the huggingface API to upload a model from huggingface to kaggle should I save it under /kaggle/input/ folder and use that as the 'Pretrained model name or path' to the model in kohya GUI?

Hhozaggers I am using kaggle and I notice the default path to the model is stabilityai/stab...

Benjamin•12/15/23, 10:20 AM

'stabilityai/stable-diffusion-xl-base-1.0' is not a local path, it points to huggingface for download

BBenjamin 'stabilityai/stable-diffusion-xl-base-1.0' is not a local path, it points to hug...

hozaggers•12/15/23, 10:21 AM

I have moy model on huggingface, wondering how I can upload it, the info I have is I need to download it from hugging face into the kaggle as a dataset, or is there a better way to do this?

Hhozaggers I have moy model on huggingface, wondering how I can upload it, the info I have ...

Benjamin•12/15/23, 10:21 AM

'stabilityai/stable-diffusion-xl-base-1.0' is the official model from stabilityai, which is hosted on huggingface

Benjamin•12/15/23, 10:23 AM

you don't need to do anything, it will be automatically downloaded and used during training

BBenjamin 'stabilityai/stable-diffusion-xl-base-1.0' is the official model from stabilitya...

hozaggers•12/15/23, 10:23 AM

I see, if I have a private model on huggingface, whats the best way to import it to kaggle if I want to reference it in the kohya GUI? do I need to upload it as a dataset in Kaggle? is there a way to get the notebook to use huggingface API to download it to kaggle directly? I want to train my private model from huggingface not the default one

Benjamin•12/15/23, 10:27 AM

if you want to use it often it makes sense to upload it as a dataset in kaggle, as you don't need to redownload it all the time. I can't recall if I'm right, but I think used kaggle datasets are also not using storage from the kaggle disk.

Benjamin•12/15/23, 10:28 AM

and then you simply copy the path from the file in your kaggle dataset and use it in 'Pretrained model name or path'

JMPAOP•12/15/23, 10:47 AM

What are json configs for text and no text encoder?

JMPAOP•12/15/23, 10:48 AM

JMPAOP•12/15/23, 10:49 AM

I’m not sure wich one use if text or no text

JJMPA Click to see attachment

Benjamin•12/15/23, 10:52 AM

he removed the ones without TE, as he suggests to always train with them

JMPAOP•12/15/23, 10:52 AM

Ok thanks

JMPAOP•12/15/23, 10:54 AM

And should I use realistic 5.1 or base 1.0 like always?

JMPAOP•12/15/23, 10:54 AM

Because always I used base 1 for realistic photos of persons

JMPAOP•12/15/23, 10:59 AM

Why he in this tutorial doesn’t use REGULARISATION images? Only training images?

Benjamin•12/15/23, 11:01 AM

if I train with other models results aren't as good as the base, I guess the training settings need to be adjusted. at least that is my experience

Benjamin•12/15/23, 11:01 AM

he always uses reg images, don't know about the video

Mnova•12/15/23, 11:25 AM

i ran into an error in deforum. does anyone know what i did wrong this time?

BBenjamin if you want to use it often it makes sense to upload it as a dataset in kaggle, ...

hozaggers•12/15/23, 1:41 PM

when I try to upload dataset to kaggle using URL I get an error saying it cannot access the file even though it is public, what am I missing?

https://huggingface.co/(username)/(modelname)/resolve/main/(modelversion).safetensors

when I run this in browser it starts automatic download. Am I doing this wrong?

hozaggers•12/15/23, 1:42 PM

(trying to upload a pre-trained model into kaggle)

Benjamin•12/15/23, 1:43 PM

never did that, I don't know. But I would rather upload it from your PC directly before digging to deep searching for an answer

hozaggers•12/15/23, 1:47 PM

hm ok, I think to upload locally it requires 2 steps, upload to kaggle working directory and then a seperate step to add it to dataset. I just noticed dataset has an option to upload via URL so seemed a more direct route

Benjamin•12/15/23, 1:47 PM

no you can directly create a dataset with an upload

Benjamin•12/15/23, 1:47 PM

and you can afterwards update datasets and upload files

Benjamin•12/15/23, 1:48 PM

just click the right button and drag and drop your model, this will create a dataset

Sshortdwarf @Dr. Furkan Gözükara downloading juggernautxl to workspace to make a LoRA on kag...

FiveSAI•12/15/23, 3:20 PM

most of the sdxl models are merges of other ones, with slight finetuning/training on them.

KGTachi•12/15/23, 3:29 PM

Hi @Dr. Furkan Gözükara what should we have as the learning rate TE for your sd1.5 json file?

Superdrew907•12/15/23, 3:51 PM

in google colab is there some way to change where you upload all your data too? anyone know?

Wax•12/15/23, 4:43 PM

hey, just joined. been trying to get the rerender installed on my machine.

BBenjamin just click the right button and drag and drop your model, this will create a dat...

hozaggers•12/15/23, 5:08 PM

I seem to be getting an error I think uploading a pre-existing model with the /kaggle/working/Kaggle_SDXL_DreamBooth_Best.json file... is that file supposed to be loaded for my pretrained models too? (all my models were trained using this json file and the kaggle tutorial). I have attatched the error where /kaggle/input/testretrain is the uploaded safetensors file.... anyone have any idea?

error.txt3.85KB

Hhozaggers I seem to be getting an error I think uploading a pre-existing model with the /k...

Benjamin•12/15/23, 5:38 PM

'/kaggle/input/testretrain' add the missing model name e.g. '/kaggle/input/testretrain/mymodelxyz.safetensors'

hozaggers•12/15/23, 5:39 PM

thanks I will try

Kaymakli Yogurt•12/15/23, 5:58 PM

Are there any AI's that can help me decide what artstyle a photo used?

Kaymakli Yogurt•12/15/23, 5:58 PM

I am trying to create images of a person with this artstyle but I am not sure which model, LORA or prompt they used to create this

Deleted User•12/15/23, 7:09 PM

Hello everyone, I encountered an error while attempting to train the "Stable Diffusion XL (SDXL) Dream Booth" model.

Deleted User•12/15/23, 7:09 PM

Deleted User•12/15/23, 7:10 PM

Does anyone know how to fix it, or how can I solve this please...

BBophedes is there a way i can fix this? i think its cuda memory. I have AMD

Furkan Gözükara SECourses•12/15/23, 10:23 PM

sadly i dont have AMD so don't know

MMnova the trainig worked, but the conversion into ckpt failed again

Furkan Gözükara SECourses•12/15/23, 10:23 PM

it is file path error

Furkan Gözükara SECourses•12/15/23, 10:24 PM

please watch this tutorial : https://youtu.be/kIyqAdd_i10

YouTubeSECourses

Stable Diffusion Google Colab, Continue, Directory, Transfer, Clone...

Our Discord : https://discord.gg/HbqgGaZVmr. This is the video where you will learn how to use Google Colab for Stable Diffusion. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on