@Dr. Furkan Gözükara how would I get it to not use VAE in addition to Juggernaut for training? Jugge
@Dr. Furkan Gözükara how would I get it to not use VAE in addition to Juggernaut for training? Juggernaut already has vae baked in
Ghello
Gcan anyone help me to download my output images, i tried using zip extention but it failed and still cost me for 2 hours of runtime, can anyone have a solution which will be less expensive or time consuming, around 100 gb of data i need to download.
Fyou dont need to provide
Fyou can use its baked
Fbest way is using hugging face
Fmethod
Flatest kaggle video showing that
FYouTubeSECourses
Master Stable Diffusion XL Training on Kaggle for Free! Welcome to this comprehensive tutorial where I'll be guiding you through the exciting world of setting up and training Stable Diffusion XL (SDXL) with Kohya on a free Kaggle account. This video is your one-stop resource for learning everything from initiating a Kaggle session with dual ...
Master Stable Diffusion XL Training on Kaggle for Free! Welcome to this comprehensive tutorial where I'll be guiding you through the exciting world of setting up and training Stable Diffusion XL (SDXL) with Kohya on a free Kaggle account. This video is your one-stop resource for learning everything from initiating a Kaggle session with dual ...Gok thnk you
F22:03 How to upload generated checkpoints / model files into Hugging Face for blazing fast upload and download
25:02 Where to find Hugging Face uploaded models after upload has been completed
25:02 Where to find Hugging Face uploaded models after upload has been completed
Dregularization pics ?
Gi am using run pod, network volume, will it work for that too?
Fit works everywhere
Fgoogle colab, your windows pc, kaggle doesnt matter
Fon runpod so easy open a juypter tab just like new terminal
Mi haven't tried two different regularization sets yet. I have trained either man or woman or style and never two different ones at a time. You can combine the trained ckpts and or loras after you train each individually.
Dhave like 20 hr left on kaggle. in your advice , i would use two folders with no reg pics
Dit will produice only one lora right , i mean i lora set(16 epochs) of two faces..
Dor else is it possible to generate couple pics and use them for regularization ?
Dif so , how will the ai know which is male and which is female ? and how would it equate it to our malke and female pics ?
They who u can insta magic animate? can u share me video? thanks
T
J@Dr. Furkan Gözükara Any videos of yours to recommend?
Im getting into the whole sdxl / stable diffusion stuff, through fooocus, and learning abotu LORAs, But seems like there is a whole lot more from dreambooth, textual params, control net, (so many ways to modify the model)
Im getting into the whole sdxl / stable diffusion stuff, through fooocus, and learning abotu LORAs, But seems like there is a whole lot more from dreambooth, textual params, control net, (so many ways to modify the model)
MI would train ohwx man in one session with man regularization images and then train again with bbuk woman with woman regularization images. When you are done you can combine the ckpts and or extract loras for each and then use the prompt I listed above. You do not need to have the man and woman together to begin with as a couple.
Gsorry, for disturbing you, but i tried using code from video in terminal it didn't worked.
Gi have set up hugging face and token but the code seems not working.
G
Gthis code i have tried.
Dsince i have trained both of our faces , i have now 16x2 loras. In this i have selected only like 3 loras for each face which are good. IN this i will select the best , next what ?
Mif you have trained loras already try the procedure I outlined above with adetailer
it takes a lot of trial and error to produce the results you want. Generate lots of images and then refine on the ones you like best
it takes a lot of trial and error to produce the results you want. Generate lots of images and then refine on the ones you like best
Dok will try it.. thanks a lot .I think in the prompt iteself i will tell which side each are .. this might help a bit in each pic
Dcheck the output folder directory again.
Dalsowhat is the error you are getting ?
FSDXL Lora with new settings, quality quite good
Git is saying
Gimpressive.
Dyou just hit play in that code line irght ?
Dwhat was the previoous commands in console ?
Dreally wish i too had similar results.. 
FI'll make a tutorial soon, i've done 600+ trainings
Gthat will be great.
FI found that dreambooth sdxl, has problems with other base models, usually only works with the base sdxl model
Feven with lora extraction
MI've had no issues with realvisxl 2 and juggernautXL (6 or 7)
AIs this the correct way to target JuggernautXL v7?
stablediffusionapi/juggernaut-xl-v7
stablediffusionapi/juggernaut-xl-v7
MIt might have to do with the training processor specs. I trained on 3060 with 12gb vram and the results were poor to average. I now train with 4090 24 gb vram or higher and the results are substantially better - amazingly actually.
Dfirst training i tried was on my 3080 10gb , and when it showed it will take 49 hours , i left and went for runpod , which finished the job around 4-5 hours.. then kaggle has been doing my jobs
FI mean like dreambooth models i've trained on the base sdxl model, worked super well with sdxl base 1.0, but the results looked weird on real vis, and jug, and others, i mean the quality is good, but the faces sometimes look off
Minteresting - i'll try another training with base 1.0 since its been awhile. thx
FMight have to do with the text encoder, not sure, but try and let me know 