How to apply densediffusion on SDXL?

[RunpuWei]

Hello, Thank you for such a wonderful job. I want to apply your densediffusion on SDXL, but I'm not sure which part needs to be changed, do you have any suggestions?

[RunpuWei]

I have modified the text encoder part, but the result is poor, as you can see, it's not as good as the positioning control on SD 1.5. It has almost no implementation features on SDXL, what might be the reason, can you give me some tips?

[YunjiKim]

There are many differences between SD and SDXL, but I believe the factor hindering the effective application of our method to SDXL is that it also takes pooled text features as input. I think this somewhat weakens our method's ability to differentiate the density of certain textual tokens according to the layout condition...

[RunpuWei]

There are many differences between SD and SDXL, but I believe the factor hindering the effective application of our method to SDXL is that it also takes pooled text features as input. I think this somewhat weakens our method's ability to differentiate the density of certain textual tokens according to the layout condition...

OK, thank you. I will continue looking for ways to implement Layout-to-Image on SDXL.

[KeyaoZhao]

There are many differences between SD and SDXL, but I believe the factor hindering the effective application of our method to SDXL is that it also takes pooled text features as input. I think this somewhat weakens our method's ability to differentiate the density of certain textual tokens according to the layout condition...

OK, thank you. I will continue looking for ways to implement Layout-to-Image on SDXL.

Hi there! I also want to try Layout-to-Image on SDXL, do you have any good idea?

[SellLin]

There are many differences between SD and SDXL, but I believe the factor hindering the effective application of our method to SDXL is that it also takes pooled text features as input. I think this somewhat weakens our method's ability to differentiate the density of certain textual tokens according to the layout condition...

OK, thank you. I will continue looking for ways to implement Layout-to-Image on SDXL.

Hi there! I also want to try Layout-to-Image on SDXL, do you have any good idea?

The Omost which implements dense diffusion-like method is compatible with sdxl, I've tried it with different sdxl models including lightning models and it works fine.

[jjiikkkk]

你们为什么不提供一个sdxl的comfyUI的工作流，你们这个功能的用户太少了