add t2i demo with ip-adapter and prior

ip_adapter/ip_adapter.py

@@ -103,11 +103,14 @@ class IPAdapter:
         ip_layers.load_state_dict(state_dict["ip_adapter"])
         
     @torch.inference_mode()
-    def get_image_embeds(self, pil_image):
+    def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
+        if pil_image is not None:
             if isinstance(pil_image, Image.Image):
                 pil_image = [pil_image]
             clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
             clip_image_embeds = self.image_encoder(clip_image.to(self.device, dtype=torch.float16)).image_embeds
+        else:
+            clip_image_embeds = clip_image_embeds.to(self.device, dtype=torch.float16)
         image_prompt_embeds = self.image_proj_model(clip_image_embeds)
         uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(clip_image_embeds))
         return image_prompt_embeds, uncond_image_prompt_embeds
@@ -119,7 +122,8 @@ class IPAdapter:
         
     def generate(
         self,
-        pil_image,
+        pil_image=None,
+        clip_image_embeds=None,
         prompt=None,
         negative_prompt=None,
         scale=1.0,
@@ -131,10 +135,13 @@ class IPAdapter:
     ):
         self.set_scale(scale)
 
+        if pil_image is not None:
             if isinstance(pil_image, Image.Image):
                 num_prompts = 1
             else:
                 num_prompts = len(pil_image)
+        else:
+            num_prompts = clip_image_embeds.size(0)
         
         if prompt is None:
             prompt = "best quality, high quality"
@@ -146,7 +153,7 @@ class IPAdapter:
         if not isinstance(negative_prompt, List):
             negative_prompt = [negative_prompt] * num_prompts
         
-        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image=pil_image, clip_image_embeds=clip_image_embeds)
         bs_embed, seq_len, _ = image_prompt_embeds.shape
         image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
         image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)