Update app.py from anycoder

app.py

@@ -1,5 +1,6 @@
 import os
 from huggingface_hub import hf_hub_download
+import spaces
 os.system("pip -qq install facenet_pytorch")
 from facenet_pytorch import MTCNN
 from torchvision import transforms
@@ -62,41 +63,53 @@ size = 256
 means = [0.485, 0.456, 0.406]
 stds = [0.229, 0.224, 0.225]
 
-t_stds = torch.tensor(stds).cuda().half()[:,None,None]
-t_means = torch.tensor(means).cuda().half()[:,None,None]
-
 img_transforms = transforms.Compose([                        
     transforms.ToTensor(),
     transforms.Normalize(means,stds)
 ])
- 
-def tensor2im(var):
-    return var.mul(t_stds).add(t_means).mul(255.).clamp(0,255).permute(1,2,0)
 
+# Load models globally (outside GPU-decorated functions)
+modelv4 = torch.jit.load(modelarcanev4).eval()
+modelv3 = torch.jit.load(modelarcanev3).eval()
+modelv2 = torch.jit.load(modelarcanev2).eval()
+
+@spaces.GPU
 def proc_pil_img(input_image, model):
+    """GPU-accelerated image processing"""
+    # Move tensors to GPU inside the decorated function
+    t_stds = torch.tensor(stds).cuda().half()[:,None,None]
+    t_means = torch.tensor(means).cuda().half()[:,None,None]
+    
+    # Move model to GPU
+    model = model.cuda().half()
+    
     transformed_image = img_transforms(input_image)[None,...].cuda().half()
+    
     with torch.no_grad():
         result_image = model(transformed_image)[0]
-        output_image = tensor2im(result_image)
+        output_image = result_image.mul(t_stds).add(t_means).mul(255.).clamp(0,255).permute(1,2,0)
         output_image = output_image.detach().cpu().numpy().astype('uint8')
         output_image = PIL.Image.fromarray(output_image)
+    
     return output_image
 
-# Load models
-modelv4 = torch.jit.load(modelarcanev4).eval().cuda().half()
-modelv3 = torch.jit.load(modelarcanev3).eval().cuda().half()
-modelv2 = torch.jit.load(modelarcanev2).eval().cuda().half()
-
+@spaces.GPU
 def process(im, version):
+    """Main processing function with GPU acceleration"""
+    if im is None:
+        return None
+    
+    # Scale image (CPU operation)
+    im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
+    
+    # Select model based on version
     if version == 'v0.4 (Recommended)':
-        im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
         res = proc_pil_img(im, modelv4)
     elif version == 'v0.3':
-        im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
         res = proc_pil_img(im, modelv3)
     else:
-        im = scale_by_face_size(im, target_face=256, max_res=1_500_000, max_upscale=1)
         res = proc_pil_img(im, modelv2)
+    
     return res
 
 # Custom theme
@@ -184,6 +197,16 @@ custom_css = """
 .example-container {
     margin-top: 1rem;
 }
+
+.gpu-badge {
+    display: inline-block;
+    background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+    color: white;
+    padding: 0.5rem 1rem;
+    border-radius: 20px;
+    font-weight: 600;
+    margin-top: 0.5rem;
+}
 """
 
 # Build the interface
@@ -196,6 +219,8 @@ with gr.Blocks() as demo:
             # 🎨 ArcaneGAN
             ### Transform Your Photos into Arcane-Style Art
             Upload a portrait and watch it transform into the stunning visual style of Netflix's Arcane series.
+            
+            <span class="gpu-badge">⚡ Powered by Zero-GPU</span>
             """
         )
         gr.Markdown(
@@ -263,6 +288,8 @@ with gr.Blocks() as demo:
             [GitHub Repository](https://github.com/Sxela/ArcaneGAN) | 
             [Original Space](https://huggingface.co/spaces/akhaliq/ArcaneGAN)
             
+            **⚡ Zero-GPU Optimization**: This Space uses Hugging Face's Zero-GPU infrastructure for efficient GPU allocation.
+            
             <div style='margin-top: 1rem;'>
                 <img src='https://visitor-badge.glitch.me/badge?page_id=akhaliq_arcanegan' alt='visitor badge'>
             </div>
@@ -274,7 +301,7 @@ with gr.Blocks() as demo:
         fn=process,
         inputs=[input_image, version_selector],
         outputs=output_image,
-        api_name="transform"
+        api_visibility="public"
     )
     
     input_image.upload(