This is my first "decent" lora and it's not perfect. I trained it on about 2000 images I picked, with automatic WD14 captions.

It works best in lower resolutions. In ComfyUI I recommend you to use clip strength between 0 and 0.1 and model strength between 0.1 and 0.6, and in A1111 between 0.1 and 0.2. Having the clip strength above 0.2 sometimes ruins the image.