Merge pull request #373 from aycaecemgul/main

repeat face embeds according to num samples to match with prompt list on portrait pipeline

ip_adapter/ip_adapter_faceid_separate.py

@@ -214,7 +214,6 @@ class IPAdapterFaceID:
     ):
         self.set_scale(scale)
 
-       
         num_prompts = faceid_embeds.size(0)
 
         if prompt is None:
@@ -224,6 +223,10 @@ class IPAdapterFaceID:
 
         if not isinstance(prompt, List):
             prompt = [prompt] * num_prompts
+        else:
+            faceid_embeds = faceid_embeds.repeat(num_samples, 1, 1)
+            num_samples = 1
+
         if not isinstance(negative_prompt, List):
             negative_prompt = [negative_prompt] * num_prompts
 
@@ -254,6 +257,7 @@ class IPAdapterFaceID:
             guidance_scale=guidance_scale,
             num_inference_steps=num_inference_steps,
             generator=generator,
+            num_images_per_prompt=num_samples,
             **kwargs,
         ).images
 
@@ -435,6 +439,10 @@ class IPAdapterFaceIDXL(IPAdapterFaceID):
 
         if not isinstance(prompt, List):
             prompt = [prompt] * num_prompts
+        else:
+            faceid_embeds = faceid_embeds.repeat(num_samples, 1, 1)
+            num_samples = 1
+
         if not isinstance(negative_prompt, List):
             negative_prompt = [negative_prompt] * num_prompts
 
@@ -470,6 +478,7 @@ class IPAdapterFaceIDXL(IPAdapterFaceID):
             negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
             num_inference_steps=num_inference_steps,
             generator=generator,
+            num_images_per_prompt=num_samples,
             **kwargs,
         ).images