Does this technique work with SDXL

[asrlhhh]

I discovered numerous hardcoded elements that restrict it to a 512x512 resolution. Additionally, with the structural modifications made to the tokenizer and text encoder in SDXL, there are now more areas that require modification. Have you verified whether the same approach is effective in SDXL?

[Matrix53]

Hello, @asrlhhh ~😊
We haven't tested our methods on SDXL. However, we believe that the attention maps should contain semantic information in SDXL. There're papers stated attention maps in SDXL and SD have similar properties.
For SDXL, we are still planning(just planning) to modify our codes. As you mention, "there are now more areas that require modification".