How to show cross attention map for checking the issue of failed compositions? about custom-diffusion HOT 3 CLOSED

Hi,

Prompt-to-prompt can be used with our diffusers-based training and models.

The original prompt-to-prompt might need to be changed for diffusers==0.14.0.
I have uploaded a working implementation here. The main changes involve the ldm_stable model load and minor modifications in register_attention_control.

Another thing that needs to be changed in prompt-to-prompt is here to the below line.

noise_pred = model.unet(latents_input, t, context)["sample"]

Let me know if you still face any issues. Thanks!

from custom-diffusion.

@nupurkmr9 Hi, thanks for you work, I'd like to ask if you have attention_map visualization codes which are not based on diffusers? Really thanks for your help!

from custom-diffusion.

Hi, I always used the diffusers based implementation for attention map visualization. Thanks.

from custom-diffusion.