Yes. I'll work on getting that fixed tomorrow with the team
Yes. I'll work on getting that fixed tomorrow with the team
Fi think he is asking any way to access jupyterlab interface from his computer
Flike gradio share
For like runpod proxy
Fadding adetailer and reactor to kaggle notebook
FKaggle notebook updated
F
F
NYes just need to double check that labs and notebook are opening the port and using host 0.0.0.0
I think notebook does automatically but need to check labs
I think notebook does automatically but need to check labs
Fnew kohya trainings just finished
Fwow took a lot
Frunpod is really really slow
Fway slower than my PC RTX 3090 for some reason
F18 different hyper parameters Kohya SS Stable Diffusion XL (SDXL) DreamBooth / FineTuning trainings just finished
After recent updates of Kohya how the…
After recent updates of Kohya how the…
Fall the test files 
now need to manually check and compare all
now need to manually check and compare all
Fty
FQuestion @Dr. Furkan Gözükara what is reccomended steps to upscale photo faces with default setting? still 50?
FSUPIR
ADoes anyone know how to fix this?
Fi use 40 steps to generate org image
Fand 70 steps 50% denoise to restore face with adetailer
Fyou are missing models
Fmy auto installers i am about to update
F
Fdoing final test
AWhat models?
F
FFip adapter
D@Dr. Furkan Gözükara I'm doing my first dreamboth training on kohya_ss previously I only used lora, I want to train many people at the same time on an xl model, friends and family from many ages and different genders, so what is the best to use dreamboth or fine tuning, one problem with dreamboth is regularization images because I'm training many classes at the same time, the other thing is the learning rate, I downloaded your preset for 24gb and the learning rate is too high compared to what I was used to use in 1.5 models that were 1e-6, I've read your YouTube post in the community that you are trying new learning rates for the new vesion of kohya, any advice?
Nfinally worked! 
thanks
thanks
Gwaiting for the best final config
Ttrying to use supir on confyui with controlnet so i can up the hallucination while keeping similar composition not sure how to connect it. i did a LatentAdd node of the canny + the supir latent and connected to the supir sampler but getting this error.
"!!! Exception during processing!!! Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!
Traceback (most recent call last):
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\comfy_extras\nodes_latent.py", line 27, in op
samples_out["samples"] = s1 + s2
"!!! Exception during processing!!! Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!
Traceback (most recent call last):
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\comfy_extras\nodes_latent.py", line 27, in op
samples_out["samples"] = s1 + s2
Tim able to do it on a regular img2img workflow with controlnet but not sure how to connect it with supir.
Fkohya need new learning rate
Fi am today gonna announce
Fwith good examples
Fdid 22 trainings
Fsadly i dont know comfyui workflow :/
Fi plan to make auto installers / downloaders for comfyui workflows
Ftime to boot up machines to start testing
F@NicB lets update my VM and set this to the configuration of Automatic1111
FF
Flets also update OneTrainer, Forge and Kohya and the Ubuntu to the latest version along with Automatic1111 Web UI
Nyep - been updating everyting and exposing jupyter notebook to the VM ip someone has
MIs there anything special about PonyXL Lora training? I'm using Prodigy and based on the parameters the Lora came out good, but it still doesn't look like the pictures in the training kit. I don't label, I use regularization images. I don't want to train DB, I want to train Lora directly.
Fmassed compute download speeds amazing
Fif working with another model, then pony not much compatiable
Fi had similar issue with realistic vision XL
MIt's okay that it's not compatible with other XL models, the Pony is very flexible and better labelled than the base model, which is why I'm thinking of exercising now, but the regular person's training is not good for some reason. That's why I'm wondering if maybe I shouldn't use the safetensor, or some other spec parameter might be needed to shape the face well.