Update ip_adapter/ip_adapter.py

ip_adapter/ip_adapter.py

@@ -203,6 +203,8 @@ class IPAdapterXL(IPAdapter):
         pil_image_5=None,
         prompt=None,
         negative_prompt=None,
+        text_scale=1.0,
+        ip_scale=1.0,
         scale_1=1.0,
         scale_2=1.0,
         scale_3=1.0,
@@ -214,8 +216,8 @@ class IPAdapterXL(IPAdapter):
         guidance_scale=7.5,
         **kwargs,
     ):
-        self.get_scale()
-        self.set_scale(scale_1)
+        #self.get_scale()
+        self.set_scale(ip_scale)
         
         if isinstance(pil_image_1, Image.Image):
             num_prompts = 1
@@ -241,26 +243,26 @@ class IPAdapterXL(IPAdapter):
         
         if pil_image_2 != None:
             print('Using secondary image.')
-            self.set_scale(scale_2)
             image_prompt_embeds_2, uncond_image_prompt_embeds_2 = self.get_image_embeds(pil_image_2)
+            image_prompt_embeds_2 = image_prompt_embeds_2 * scale_2
             image_prompt_embeds_list.append(image_prompt_embeds_2)
             uncond_image_prompt_embeds_list.append(uncond_image_prompt_embeds_2)
         if pil_image_3 != None:
-            print('Using secondary image.')
-            self.set_scale(scale_3)
+            print('Using tertiary image.')
             image_prompt_embeds_3, uncond_image_prompt_embeds_3 = self.get_image_embeds(pil_image_3)
+            image_prompt_embeds_3 = image_prompt_embeds_3 * scale_3
             image_prompt_embeds_list.append(image_prompt_embeds_3)
             uncond_image_prompt_embeds_list.append(uncond_image_prompt_embeds_3)
         if pil_image_4 != None:
             print('Using secondary image.')
-            self.set_scale(scale_4)
             image_prompt_embeds_4, uncond_image_prompt_embeds_4 = self.get_image_embeds(pil_image_4)
+            image_prompt_embeds_4 = image_prompt_embeds_4 * scale_4
             image_prompt_embeds_list.append(image_prompt_embeds_4)
             uncond_image_prompt_embeds_list.append(uncond_image_prompt_embeds_4)
         if pil_image_5 != None:
             print('Using secondary image.')
-            self.set_scale(scale_5)
             image_prompt_embeds_5, uncond_image_prompt_embeds_5 = self.get_image_embeds(pil_image_5)
+            image_prompt_embeds_5 = image_prompt_embeds_5 * scale_5
             image_prompt_embeds_list.append(image_prompt_embeds_5)
             uncond_image_prompt_embeds_list.append(uncond_image_prompt_embeds_5)
 
@@ -276,7 +278,8 @@ class IPAdapterXL(IPAdapter):
         with torch.inference_mode():
             prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = self.pipe.encode_prompt(
                 prompt, num_images_per_prompt=num_samples, do_classifier_free_guidance=True, negative_prompt=negative_prompt)
-            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            prompt_embeds = prompt_embeds * text_scale
+            prompt_embeds = prompt_embeds * text_scale = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
             negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
             
         generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None