Optimize CPU performance: reduce inference steps and guidance scale for faster processing

app/colorize_model.py

@@ -111,18 +111,23 @@ class ColorizeModel:
         
         return image
     
-    def colorize(self, image: Image.Image, num_inference_steps: int = 20) -> Image.Image:
+    def colorize(self, image: Image.Image, num_inference_steps: int = None) -> Image.Image:
         """
         Colorize a grayscale image
         
         Args:
             image: PIL Image (grayscale or color)
-            num_inference_steps: Number of inference steps
+            num_inference_steps: Number of inference steps (auto-adjusted for CPU/GPU)
             
         Returns:
             Colorized PIL Image
         """
         try:
+            # Optimize inference steps based on device
+            if num_inference_steps is None:
+                # Use fewer steps on CPU for faster processing
+                num_inference_steps = 8 if self.device == "cpu" else 20
+            
             # Preprocess image
             control_image = self.preprocess_image(image)
             original_size = image.size
@@ -131,6 +136,9 @@ class ColorizeModel:
             prompt = "colorize this black and white image, high quality, detailed, vibrant colors, natural colors"
             negative_prompt = "black and white, grayscale, monochrome, low quality, blurry, desaturated"
             
+            # Adjust guidance scale for CPU (lower = faster)
+            guidance_scale = 5.0 if self.device == "cpu" else 7.5
+            
             # Generate colorized image based on model type
             if self.model_type == "controlnet":
                 # Use ControlNet pipeline
@@ -139,7 +147,7 @@ class ColorizeModel:
                     image=control_image,
                     negative_prompt=negative_prompt,
                     num_inference_steps=num_inference_steps,
-                    guidance_scale=7.5,
+                    guidance_scale=guidance_scale,
                     controlnet_conditioning_scale=1.0,
                     generator=torch.Generator(device=self.device).manual_seed(42)
                 )