How to Generate added_cond_kwargs for SDXL in demo_image_editing.ipynb with xl_sliders/winter_weather.pt?

[JasonBourne1998]

Thank you for providing such an excellent project!

I am currently experimenting with demo_image_editing.ipynb and trying to adapt it to work with SDXL instead of the default model. Specifically, I am using the xl_sliders/winter_weather.pt weights. However, I am uncertain about how to correctly generate the added_cond_kwargs to pass into self.model.unet.

Could you kindly provide guidance or an example of how to prepare the added_cond_kwargs for SDXL in this context? For instance:
• How should add_time_ids and text_embeds be generated and shaped for compatibility with SDXL?
• Are there specific differences in the way SDXL expects these inputs compared to previous models?

Additionally, if you already have a demo or notebook (e.g., an equivalent to demo_image_editing.ipynb) tailored for SDXL sliders, I would greatly appreciate it if you could point me toward that resource.

[JasonBourne1998]

@rohitgandikota Could you plz help on this?

[rohitgandikota]

Hi @JasonBourne1998

This is an open issue in the repo. This leads back to having an implementation open-sourced for null-inversion for SDXL (different work by team from Google). This is essential for that to work. We are still looking for people to work on this.