Update pipeline_stable_diffusion_3_ipa.py

pipeline_stable_diffusion_3_ipa.py

@@ -1161,7 +1161,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_image_embeds_2 = clip_image_embeds_2.to(device, dtype=dtype)
             clip_image_embeds_2 = self.image_encoder(clip_image_embeds_2, output_hidden_states=True).hidden_states[-2]
             clip_image_embeds_2 = clip_image_embeds_2 * scale_2
-            image_prompt_embeds_list.append(image_prompt_embeds_2)
+            image_prompt_embeds_list.append(clip_image_embeds_2)
         if clip_image_3 != None:
             print('Using tertiary image.')
             clip_image_3 = clip_image_3.resize((max(clip_image_3.size), max(clip_image_3.size)))
@@ -1170,7 +1170,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_image_embeds_3 = clip_image_embeds_3.to(device, dtype=dtype)
             clip_image_embeds_3 = self.image_encoder(clip_image_embeds_3, output_hidden_states=True).hidden_states[-2]
             clip_image_embeds_3 = clip_image_embeds_3 * scale_3
-            image_prompt_embeds_list.append(image_prompt_embeds_3)
+            image_prompt_embeds_list.append(clip_image_embeds_3)
         if clip_image_4 != None:
             print('Using quaternary image.')
             clip_image_4 = clip_image_4.resize((max(clip_image_4.size), max(clip_image_4.size)))
@@ -1179,7 +1179,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_image_embeds_4 = clip_image_embeds_4.to(device, dtype=dtype)
             clip_image_embeds_2 = self.image_encoder(clip_image_embeds_4, output_hidden_states=True).hidden_states[-2]
             clip_image_embeds_4 = clip_image_embeds_4 * scale_4
-            image_prompt_embeds_list.append(image_prompt_embeds_4)
+            image_prompt_embeds_list.append(clip_image_embeds_4)
         if clip_image_5 != None:
             print('Using quinary image.')
             clip_image_5 = clip_image_5.resize((max(clip_image_5.size), max(clip_image_5.size)))
@@ -1188,7 +1188,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_image_embeds_5 = clip_image_embeds_5.to(device, dtype=dtype)
             clip_image_embeds_5 = self.image_encoder(clip_image_embeds_5, output_hidden_states=True).hidden_states[-2]
             clip_image_embeds_5 = clip_image_embeds_5 * scale_5
-            image_prompt_embeds_list.append(image_prompt_embeds_5)
+            image_prompt_embeds_list.append(clip_image_embeds_5)
             
         # Concatenate the image embeddings
        ## clip_image_embeds = torch.mean(torch.stack(image_prompt_embeds_list), dim=0)