Update pipeline_stable_diffusion_3_ipa.py

pipeline_stable_diffusion_3_ipa.py

@@ -922,8 +922,9 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
 
     @torch.inference_mode()
     def encode_clip_image_emb(self, clip_image, device, dtype):
-
-        # clip
+        if isinstance(clip_image, Image.Image):
+            clip_image = [clip_image]
+            # clip
         clip_image_tensor = self.clip_image_processor(images=clip_image, return_tensors="pt").pixel_values
         clip_image_tensor = clip_image_tensor.to(device, dtype=dtype)
         clip_image_embeds = self.image_encoder(clip_image_tensor, output_hidden_states=True).hidden_states[-2]
@@ -1177,7 +1178,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
 
         # Create a linear layer
         embedding_dim = concatenated_embeds.shape[-1]  # Get the embedding dimension
-        linear_layer = nn.Linear(embedding_dim * len(image_prompt_embeds_list), embedding_dim, dtype=torch.bfloat16).to('cuda')
+        linear_layer = nn.Linear(len(image_prompt_embeds_list) * embedding_dim, embedding_dim, dtype=torch.bfloat16).to('cuda')
         # Move `concatenated_embeds` to the GPU if it's on the CPU
         '''
         if concatenated_embeds.device.type == 'cpu':