AP123 commited on
Commit
f32daf2
1 Parent(s): 87af0cd

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +10 -9
app.py CHANGED
@@ -43,29 +43,30 @@ def transform_image(face_image):
43
  else:
44
  raise ValueError("Unsupported image format")
45
 
46
- # Resize the face image
47
  processed_face_image = processed_face_image.resize(desired_size, Image.LANCZOS)
48
-
49
- # Convert PIL images to PyTorch tensors
50
  processed_face_tensor = transforms.ToTensor()(processed_face_image).unsqueeze(0).to("cuda")
51
- style_image_tensor = transforms.ToTensor()(style_image).unsqueeze(0).to("cuda")
52
 
53
- # Ensure tensors are the correct shape (C, H, W)
54
- if processed_face_tensor.shape[1:] != (3, 1280, 1280):
55
- raise ValueError(f"Face image tensor shape is {processed_face_tensor.shape}, but expected shape is (3, 1280, 1280)")
 
56
 
57
  # Perform the transformation using the configured pipeline
58
  image = pipeline(
59
  prompt="soyjak",
60
- ip_adapter_image=[style_image_tensor, processed_face_tensor],
61
  negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
62
  num_inference_steps=30,
63
  generator=generator,
64
  ).images[0]
65
 
 
 
 
66
  # Move the pipeline back to CPU after processing to release GPU resources
67
  pipeline.to("cpu")
68
- return transforms.ToPILImage()(image.squeeze(0))
69
 
70
  # Gradio interface setup
71
  demo = gr.Interface(
 
43
  else:
44
  raise ValueError("Unsupported image format")
45
 
46
+ # Resize the face image and convert to tensor
47
  processed_face_image = processed_face_image.resize(desired_size, Image.LANCZOS)
 
 
48
  processed_face_tensor = transforms.ToTensor()(processed_face_image).unsqueeze(0).to("cuda")
 
49
 
50
+ # Load the style image from the local path, resize it and convert to tensor
51
+ style_image_path = "examples/soyjak2.jpeg" # Ensure this path is correct
52
+ style_image = Image.open(style_image_path).resize(desired_size, Image.LANCZOS)
53
+ style_image_tensor = transforms.ToTensor()(style_image).unsqueeze(0).to("cuda")
54
 
55
  # Perform the transformation using the configured pipeline
56
  image = pipeline(
57
  prompt="soyjak",
58
+ ip_adapter_image=[style_image_tensor, processed_face_tensor], # Ensure these are tensors
59
  negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
60
  num_inference_steps=30,
61
  generator=generator,
62
  ).images[0]
63
 
64
+ # Convert the tensor to a PIL Image to display it in Gradio
65
+ image = transforms.ToPILImage()(image.squeeze(0))
66
+
67
  # Move the pipeline back to CPU after processing to release GPU resources
68
  pipeline.to("cpu")
69
+ return image
70
 
71
  # Gradio interface setup
72
  demo = gr.Interface(