Presently works best with an updated comfyui. Put in the embedding folder. For me it worked well with euler sampler.
At the time of making this model it doesn't work with Stable Diffusion web UI.
Description
FAQ
Comments (8)
How did you perform the training? What script?
https://github.com/bmaltais/kohya_ss
I can share my setting if you need them.
@summer4love263 Thanks for your impressive work, I want to know your setting for train.
@Number18_tong
can download the setting from here. Bit of trail and error. This does work but you may want to tweak it.
https://github.com/bluntstone/kohya_ss_setting_ti
My ComfyUI stopped working, using linux. I have reinstalled it and it refuses to run. Not sure whats wrong. So until I can get it running I stopped making embedding for sdxl because I can't test them. Waiting for a new version.
@summer4love263 Thanks a lot
@summer4love263 I have tested for the GiseleBündchen.safetensors in diffusers' pipeline, and the result is real good. How many train images for the GiseleBündchen.safetensors, and what is the image size, Please?
@Number18_tong Been using 50 images of late. I usually make it a max 1024 either width or height. It doesn't have to be square. I don't think you have to resize, sometimes I don't, it just I like to use inpaint in sd and it is slow with large images. I use it to clean them up, text. Also of late I stopped using text files. Just images only. https://civitai.com/models/141234/caroline-winberg model was made with no text files. I find with sdxl it just works better. I think you need a 3090 or a 4090 for larger images. Not totally sure. Keep batch size to 1 so you don't run out of memory.
@summer4love263 Thanks a million. I really appreciate it !!