I'll look into bad prompt, but I already have easynegative.
Using Embeddings though, didn't make a difference at all. I don't know if I got unlucky and the models I tested it on already had easynegative built in or what.
How do I apply it, then? I click the Add LORA button, then select the model. The thing says I have the right keyword entered into the box, and then the final product has a new button called Textual Inversions with the Embedding I chose. But the results don't change.
@Cipiteca396 that should be it. I'm not sure why it'd be the same. maybe for bad_prompt? I couldn't find it on happyaccidents so the keyword probably does nothing
Changing the prompt will always change the output. The Embedding itself is unnecessary. Adding it in is an extra step that doesn't change the result at all as far as I've observed.
Aha, it did work with the baddream embedding you used (when it was used on Dreamshaper ).
So easynegative really must be built into most of the models I use, no embeddings needed. Or that particular embedding is just broken.