Switch to PIL image processing

handler.py

@@ -1,7 +1,7 @@
 from typing import  Dict, List, Any
 import torch
-from torch import autocast
 from diffusers import StableDiffusionPipeline, EulerAncestralDiscreteScheduler
+from PIL import Image
 import base64
 from io import BytesIO
 
@@ -28,32 +28,39 @@ class EndpointHandler():
             A :obj:`dict`:. base64 encoded image
         """
         inputs = data.pop("inputs", data)
+        params = data.pop("parameters", data)
 
         # hyperparamters
-        num_inference_steps = data.pop("num_inference_steps", 25)
-        guidance_scale = data.pop("guidance_scale", 7.5)
-        negative_prompt = data.pop("negative_prompt", None)
-        height = data.pop("height", None)
-        width = data.pop("width", None)
-        manual_seed = data.pop("manual_seed", -1)
+        num_inference_steps = params.pop("num_inference_steps", 20)
+        guidance_scale = params.pop("guidance_scale", 7.5)
+        negative_prompt = params.pop("negative_prompt", None)
+        height = params.pop("height", None)
+        width = params.pop("width", None)
+        manual_seed = params.pop("manual_seed", -1)
 
         generator = torch.Generator(device).manual_seed(manual_seed)
 
+        if encoded_image is not None:
+            image = self.decode_base64_image(encoded_image)
+
         # run inference pipeline
-        with autocast(device.type):
-            image = self.pipe(inputs,
-                        generator=generator,
+        out = self.pipe(inputs, 
+                        image=image,             
+                        generator=generator,             
                         num_inference_steps=num_inference_steps,
                         guidance_scale=guidance_scale,
                         num_images_per_prompt=1,
                         negative_prompt=negative_prompt,
                         height=height,
-                        width=width).images[0]  
+                        width=width
+        )
             
-        # encode image as base 64
-        buffered = BytesIO()
-        image.save(buffered, format="JPEG")
-        img_str = base64.b64encode(buffered.getvalue())
-
-        # postprocess the prediction
-        return {"image": img_str.decode()}
+        # return first generate PIL image
+        return out.images[0]
+    
+    # helper to decode input image
+    def decode_base64_image(self, image_string):
+        base64_image = base64.b64decode(image_string)
+        buffer = BytesIO(base64_image)
+        image = Image.open(buffer)
+        return image