How about global properties?
KIMGEONUNG opened this issue · 2 comments
Hi, Thank you for your amazing work, and I have a question about your research, not code.
The paper focused on object-centric words such as elephant or crown.
However, words indicating the global property, e.g., "at night", are also frequently omitted in the ouputs of T2I diffusion models. So, I just wonder whether your method also works to the words representing global property.
If not, I also want to know your opinion for that reasons.
Thank you.
Hi @KIMGEONUNG, thanks for your interest!
In the paper, we present some examples with global properties such as “a Picasso painting in a garden” and “A cat and a dog reading in a library” so the method works on background subjects as well. I hope this answers your question.
Here’s an example I got with a random seed for “a cat and a dog at night”:
