Update api/ltx_server_refactored.py

api/ltx_server_refactored.py

@@ -93,55 +93,6 @@ from ltx_video.utils.skip_layer_strategy import SkipLayerStrategy
 import ltx_video.pipelines.crf_compressor as crf_compressor
 
 
-
-def load_image_to_tensor_with_resize_and_crop(
-    image_input: Union[str, Image.Image],
-    target_height: int = 512,
-    target_width: int = 768,
-    just_crop: bool = False,
-) -> torch.Tensor:
-    """Load and process an image into a tensor.
-
-    Args:
-        image_input: Either a file path (str) or a PIL Image object
-        target_height: Desired height of output tensor
-        target_width: Desired width of output tensor
-        just_crop: If True, only crop the image to the target size without resizing
-    """
-    if isinstance(image_input, str):
-        image = Image.open(image_input).convert("RGB")
-    elif isinstance(image_input, Image.Image):
-        image = image_input
-    else:
-        raise ValueError("image_input must be either a file path or a PIL Image object")
-
-    input_width, input_height = image.size
-    aspect_ratio_target = target_width / target_height
-    aspect_ratio_frame = input_width / input_height
-    if aspect_ratio_frame > aspect_ratio_target:
-        new_width = int(input_height * aspect_ratio_target)
-        new_height = input_height
-        x_start = (input_width - new_width) // 2
-        y_start = 0
-    else:
-        new_width = input_width
-        new_height = int(input_width / aspect_ratio_target)
-        x_start = 0
-        y_start = (input_height - new_height) // 2
-
-    image = image.crop((x_start, y_start, x_start + new_width, y_start + new_height))
-    if not just_crop:
-        image = image.resize((target_width, target_height))
-
-    image = np.array(image)
-    image = cv2.GaussianBlur(image, (3, 3), 0)
-    frame_tensor = torch.from_numpy(image).float()
-    frame_tensor = crf_compressor.compress(frame_tensor / 255.0) * 255.0
-    frame_tensor = frame_tensor.permute(2, 0, 1)
-    frame_tensor = (frame_tensor / 127.5) - 1.0
-    # Create 5D tensor: (batch_size=1, channels=3, num_frames=1, height, width)
-    return frame_tensor.unsqueeze(0).unsqueeze(2)
-
 def create_latent_upsampler(latent_upsampler_model_path: str, device: str):
     latent_upsampler = LatentUpsampler.from_pretrained(latent_upsampler_model_path)
     latent_upsampler.to(device)
@@ -303,6 +254,56 @@ class VideoService:
     # --- Métodos Públicos (API do Serviço) ---
     # --------------------------------------------------------------------------
 
+    def _load_image_to_tensor_with_resize_and_crop(
+        self,
+        image_input: Union[str, Image.Image],
+        target_height: int = 512,
+        target_width: int = 768,
+        just_crop: bool = False,
+    ) -> torch.Tensor:
+        """Load and process an image into a tensor.
+    
+        Args:
+            image_input: Either a file path (str) or a PIL Image object
+            target_height: Desired height of output tensor
+            target_width: Desired width of output tensor
+            just_crop: If True, only crop the image to the target size without resizing
+        """
+        if isinstance(image_input, str):
+            image = Image.open(image_input).convert("RGB")
+        elif isinstance(image_input, Image.Image):
+            image = image_input
+        else:
+            raise ValueError("image_input must be either a file path or a PIL Image object")
+    
+        input_width, input_height = image.size
+        aspect_ratio_target = target_width / target_height
+        aspect_ratio_frame = input_width / input_height
+        if aspect_ratio_frame > aspect_ratio_target:
+            new_width = int(input_height * aspect_ratio_target)
+            new_height = input_height
+            x_start = (input_width - new_width) // 2
+            y_start = 0
+        else:
+            new_width = input_width
+            new_height = int(input_width / aspect_ratio_target)
+            x_start = 0
+            y_start = (input_height - new_height) // 2
+    
+        image = image.crop((x_start, y_start, x_start + new_width, y_start + new_height))
+        if not just_crop:
+            image = image.resize((target_width, target_height))
+    
+        image = np.array(image)
+        image = cv2.GaussianBlur(image, (3, 3), 0)
+        frame_tensor = torch.from_numpy(image).float()
+        frame_tensor = crf_compressor.compress(frame_tensor / 255.0) * 255.0
+        frame_tensor = frame_tensor.permute(2, 0, 1)
+        frame_tensor = (frame_tensor / 127.5) - 1.0
+        # Create 5D tensor: (batch_size=1, channels=3, num_frames=1, height, width)
+        return frame_tensor.unsqueeze(0).unsqueeze(2)
+
+
     
     def generate_low_resolution1(self, prompt: str, negative_prompt: str, height: int, width: int, duration_secs: float, guidance_scale: float, seed: Optional[int] = None, conditioning_items: Optional[List[PatchedConditioningItem]] = None) -> Tuple[str, str, int]:
         """
@@ -698,7 +699,7 @@ class VideoService:
 
     def _prepare_conditioning_tensor_from_path(self, filepath: str, height: int, width: int, padding: Tuple) -> torch.Tensor:
         """Carrega uma imagem, redimensiona, aplica padding e move para o dispositivo."""
-        tensor = load_image_to_tensor_with_resize_and_crop(filepath, height, width)
+        tensor = self._load_image_to_tensor_with_resize_and_crop(filepath, height, width)
         tensor = F.pad(tensor, padding)
         return tensor.to(self.device, dtype=self.runtime_autocast_dtype)