ASK FOR prompt_guard

[ngocdai101004]

Hello, I’m currently studying methods for content control in images generated by generative models, and I’m glad to see that your paper/source code is available. However, I’m having some issues with the source code.

I’d like to ask about the parameter

--placeholder_token="<prompt_guard>"

Is there any standard or rule for what prompt_guard should be? Should it be a sentence or a single word?

Also, for a pretrained model, the variable

token_list = [token1, token2, ...]  # the corresponding tokens of your embeddings

— what exactly are these tokens referring to?

[lingzhiyxp]

Hello! Thanks for using our code. The "<prompt_guard>" is just a place holder. This will add a new token in the vocabulary in the text encoder, which represents the token corresponding to our additional soft embedding.
For the second question, [token1, token2, ...] is the same variable with token1 = "<prompt_guard_1>" token2 = "<prompt_guard_2>", which is just as what I explained above.