Kudos to nice model and one question)
Great model already! Is it possible to share setting used for a training? Just settings - no datasets or anything. If no then no - just asking)
well, mostly i’m using Kohya’s default parameters, but in this model, i’m using 1-e6 LR, no mixed precision (since at that time, Fp16 VAE doesn’t exist), that’s all.
and i’m using BLIP to caption, but i think WD tagger is much better, if you want to use SDXL Base 0.9 as pretrained model. BLIP probably conflicting with SDXL tags
i really like this model too. i'm using Comfyui to generate images. your other model notwaifu-diffusion-xl does not produce the same quality😓
Thanks for info! Appreciated)
i really like this model too. i'm using Comfyui to generate images. your other model notwaifu-diffusion-xl does not produce the same quality😓
you need to use WD aesthetics tags to produce good result with notwaifu-diffusion.