Recently I have some views, please let me know ,if is there any problems
Recently I have some views, please let me know ,if is there any problems
H1.check unfreeze model will be better to help the model learning instance images datas, can also improve the generalization ability of the training model
H- When training design style model, class images are not necessary ,especially the instance images amount above 100.
H- Writing class name of class images ，leaving class images empty will also be helpful
H4.the function of class images is to help the model transfer learning
RSorted
H5.class name is to avoid pollution of the other class of the base model
H- If not well prepared, the class image can also mislead or pollute the training data
H@Furkan Gözükara SECourses
Adoes anybody know if the open pose editor extention is broken? When I make one and send to txt2img nothing happens, I jusdt get a dark screen.. using controlnet 1.1
Ai encounter this problem
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when
checking argument for argument mat2 in method wrapper_CUDA_mm)
how can i fix it
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when
checking argument for argument mat2 in method wrapper_CUDA_mm)
how can i fix it
Ai have rtx3050 and rtx 3080 in pc. my default gpu is rtx 3080
I have tried
I have tried
- reinstall
- install Python 3.10.9
- set cuda visible devices system 0,1
But problem is still.
Fi would try restart computer
Fclass images to prevent overtraining and keeping sanity of the model. if i dont use class images, and do some extensive training, when i type man i would see myself :d
Fcould be. they are still doing updates i think
Fcan you show your startup command line arguments?
Fhere example of mine
F
Vhey, is there any way to download a model from huggingface directly to google drive using colab?
Fuse !wget command
Fi am working on a very unique tutorial
Fthat no other youtuber has ever covered 
Vi tried to use it, but turns out the huggingface link doens't have the api link in it like civitai does, is there a workaround for this?
Voh?? what is it about?!
Fthere is link
F
Fit is using AI to get best images out of thousands of images
Fvery useful
Fi have prepared 13 new images of my self
Fgonna do train now
Vohhhh, thank you! i was putting in the url for the download page lol, im dumb
Vnice, that would be very useful!
Vlets gooo, looking forward to the video!
Fthanks
Fi really like sd dreambooth than kohya
F
Feverything so clear
Vi haven't used either lol
МHi all. Google translator.
- Positive prompts have a token limit, after which the accuracy decreases. Is there such a limitation for negative prompts?
- When typing a certain "critical mass" at which the accuracy of the request decreases, are positive and negative tokens considered together, or separately?
AError when generating images with the Composable LoRA extension on AUTO1111 fork
https://github.com/vladmandic/automatic/issues/629
https://github.com/vladmandic/automatic/issues/629
GitHub
Issue Description Description: When generating images after installing the Composable LoRA extension (https://github.com/opparco/stable-diffusion-webui-composable-lora) on the AUTO1111 fork, an err...
A
Ftoken limit is 75 . when more tokens are used their weights are averaged
Fnegative tokens are used to derive away image from those latent space. they are used at a certain step in the architecture of the model
Fye not all lora supported atm
Fi am also comparing dreambooth and kohya lora atm
Zthanks for the classification images doc
Zcan we run deep floyd locally on 11 gigs of vram?
Hi tried not using classification dataset and class image name in lora, the result is not bad
H120 instance images
Hwhat is the optimal value of the train unet (steps ratio of the textencoder ) for train a style /person