Been testing this. Use in the negative prompt. I been using it for portraits.
Not sure if I made it correctly. I didn't find any info for making negative Textual inversions.
This is how I made it:
I put all the negative prompts in the prompt and rendered about 100 images and then trained a textual inversion with them. I didn't use blip, trained directly with the images.
If you thing gross is a bad name for it, just rename the file and use whatever name you give it.