Update app.py

app.py

@@ -2,6 +2,7 @@ import gradio as gr
 import torch
 from PIL import Image
 from diffusers import AutoPipelineForText2Image, DDIMScheduler
+from transformers import CLIPVisionModelWithProjection
 import numpy as np
 import spaces  # Make sure to import spaces
 
@@ -25,11 +26,14 @@ pipeline.load_ip_adapter(
 )
 pipeline.set_ip_adapter_scale([0.7, 0.5])
 
-# Decorate the transform_image function to run on GPU
+# Ensure the model and its components are moved to GPU
+pipeline.to("cuda")
+
+# Define the desired size
+desired_size = (1024, 1024)
+
 @spaces.GPU
 def transform_image(face_image):
-    # Move the pipeline to GPU inside the function
-    pipeline.to("cuda")
     generator = torch.Generator(device="cuda").manual_seed(0)
 
     # Process the input face image
@@ -40,16 +44,19 @@ def transform_image(face_image):
     else:
         raise ValueError("Unsupported image format")
 
-    # Load the style image from the local path
-    style_image_path = "./examples/soyjak2.jpeg"
-    style_image = Image.open(style_image_path)
+    # Resize the face image
+    processed_face_image = processed_face_image.resize(desired_size, Image.LANCZOS)
+
+    # Load and resize the style image from the local path
+    style_image_path = "examples/soyjak2.jpeg"
+    style_image = Image.open(style_image_path).resize(desired_size, Image.LANCZOS)
 
     # Perform the transformation using the configured pipeline
     image = pipeline(
         prompt="soyjak",
         ip_adapter_image=[style_image, processed_face_image],
         negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
-        num_inference_steps=50,
+        num_inference_steps=30,
         generator=generator,
     ).images[0]