This lora works, but I'm getting some blurry results. Looking through the tags, I think it's because you tagged every image identically, with none of the trigger words actually being present in the dataset apart from "green" and "dress".

I think they used to call that the "brute force method", but they'd usually get a bunch of images and tie it to a trigger word rather than force the neural network to weight several tags identically.

Just don't forget to tag things differently for differences between each image in the dataset. If you are having trouble tagging for the dress itself, just make a trigger word (in the actual dataset) for the whole thing, or at least the parts you can not describe.

Apparently this doesnt work very well with the fluttershy lora