Update pipeline_stable_diffusion_3_ipa.py

pipeline_stable_diffusion_3_ipa.py

@@ -1204,15 +1204,13 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         #clip_image_embeds = torch.cat([torch.zeros_like(torch.stack(image_prompt_embeds_list)), torch.stack(image_prompt_embeds_list)], dim=0).mean(dim=0)
         # FAILS clip_image_embeds = torch.cat(torch.stack(image_prompt_embeds_list), dim=0).mean(dim=0)
         # FAILS TIMESTEPS clip_image_embeds = torch.cat(image_prompt_embeds_list, dim=0).mean(dim=0)
-        clip_image_embeds_cat_list = torch.cat(image_prompt_embeds_list).mean(dim=0)
-        print('catted with mean shape 1: ', clip_image_embeds_cat_list.shape)
-        clip_image_embeds_stack_list = torch.stack(image_prompt_embeds_list).mean(dim=0)
-        print('stacked with mean shape 1: ', clip_image_embeds_stack_list.shape)
-        clip_image_embeds_cat_list = torch.cat(image_prompt_embeds_list)
-        print('catted without mean shape 1: ', clip_image_embeds_cat_list.shape)
-        clip_image_embeds_stack_list = torch.stack(image_prompt_embeds_list)
-        print('stacked without mean shape 1: ', clip_image_embeds_stack_list.shape)
 
+        clip_image_embeds_stack_list = torch.stack(image_prompt_embeds_list).mean(dim=0)
+        print('stacked with mean dim 0 shape: ', clip_image_embeds_stack_list.shape)
+        zeroes_tensor = torch.zeros_like(clip_image_embeds_stack_list)
+        print('zeros shape: ', stacked_image_embeds.shape)
+        clip_image_embeds = torch.cat([zeroes_tensor, clip_image_embeds_stack_list], dim=0)
+        print('embeds shape: ', clip_image_embeds.shape)
         
         # 1. Stack the image embeddings
         stacked_image_embeds = torch.cat(image_prompt_embeds_list, dim=1)
@@ -1236,7 +1234,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         print('shape 4a: ', clip_image_embeds_repeat_cat.shape)
         clip_image_embeds_repeat_cat_1 = torch.cat([zeros_tensor_repeat, clip_image_embeds_repeat], dim=1)
         print('shape 4b: ', clip_image_embeds_repeat_cat_1.shape)
-        clip_image_embeds = clip_image_embeds_repeat_cat
+        #clip_image_embeds = clip_image_embeds_repeat_cat
         '''
         #clip_image_embeds = clip_image_embeds.unsqueeze(0) # Add a dimension at the beginning so now you have [1, 2*seq_len_img, embed_dim_img]
         print('shape 5: ', clip_image_embeds.shape)