Hi, I noticed in the "Training dataset tags" in Forge UI that you included parentheses in the captions for your training data. Since parentheses are used to increase the weight of tokens in the prompt, it’s best to avoid using those in the training captions.

For instance, one of the trained tokens ended up as "dress (white top" and another as "black botton)". The comma between these phrases caused them to split into two separate tokens. (Also, "botton" should probably be "bottom," in case you do a version 2.)

Just a heads up to avoid unintended behavior in image generation.