Hello I am trying to get my images trained in dreambooth colab notebook but I don't get very good results importing the model in automatic1111 as I did with old dreambooth notebooks using compvis stable diffusion 1.4 model (if I use the original mdjrny-v4.ckpt the images are vey good, also if not very precise with the prompt). Are there any recommandations to train this model with new images? I think the midjouney style looks very nice and getting the images I choose in the result would be good.
I just want say that is possible to train your images with this model throught fast dreambooth using the right configuration. One important thing is to uncheck fp16 because all outputs will come out glitched