with 12 gb you can do SDXL lora
with 12 gb you can do SDXL lora
Fbest option for you
Fsd 1.5 dreambooth perhaps with optimization but quality i dont know need testing
Fcan you show your settings? configuration
Ldo you think that i will get bettwe result with a 24gb gpu on runpod?
F100%
Fwith dreambooth of SDXL
FLi dont know how to use SDXL, thats why im training for sd 1.5
Ldo you have a video how to train model with dreambooth for sdxl?
Fi have for lora
Fyou only need to load config
Ffrom patreon
Fand prepare dataset nothing else
FYouTubeSECourses
If you don't have a GPU, or have a strong GPU, or you are using Mac and your computer not supporting Stable Diffusion training, SDXL training, then this is the tutorial you are looking for. In this tutorial, we will use a cheap cloud GPU service provider RunPod to use both Stable Diffusion Web UI Automatic1111 and Stable Diffusion trainer Kohya ...
Fafter watching this video you can load config and use it
Fi will also hopefully record quick video for patreon
Fcurrently fixing kaggle notebook
Lis ti still possible to train model with dreambooth?
Fyes
S@Dr. Furkan Gözükara I’m using runpod what disk space should I use for training 100GB for both? persistence and the other one?
Fwith lora
Fyou need only volume disk
Fnot the other one
Lcan you please have a look at my dataset and give some tips?
Sthe volume disk 100gb and the other one could be 20?
Fsure pm me
Fyep
Ldone
SI’m having double bodies/faces following your training tutorial also naked images using real vision the same parameters as yours
Scanvas 768x1024 I’m also using the data images of woman. any advise?
EAdetailer used here? I prefer resaults without adetailer as it keeps the original texutre and style
Fthose models are nasty
Fi think i would try sdxl
Fadetailer were used for realistic images
Ffor styled i didnt use
Eohh and not used for stylized, very good idea
Fyes
Ealso should we then use Text Encoder training or not? kept the suspence going on that twit 
Elooks pretty identical
Fthis one is amazing
Fi think looks like not very necessary
Fbut can be used too :d
R@Dr. Furkan Gözükara does NOT-raining text encoder save memory and speed? Is it worth turning off TE learning?
ROh just noticed someone asked the same question
Fit is turned off by default
Fit increases vram slighty and reduces training speed slightly
RAlso, where do you explain the block training --block_lr 1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5
EI had Very good success using control net Depth, Canny and openpose to upscale and bring out details in an old photo from this
