diff --git a/src/diffusers/image_processor.py b/src/diffusers/image_processor.py
index 4f6f4bd52b9c..e69f6c3a40ff 100644
--- a/src/diffusers/image_processor.py
+++ b/src/diffusers/image_processor.py
@@ -1048,28 +1048,16 @@ def rgblike_to_depthmap(image: np.ndarray | torch.Tensor) -> np.ndarray | torch.
         #    for return value from a library function.
 
         if isinstance(image, torch.Tensor):
-            # Cast to a safe dtype (e.g., int32 or int64) for the calculation
-            original_dtype = image.dtype
             image_safe = image.to(torch.int32)
-
-            # Calculate the depth map
             depth_map = image_safe[:, :, 1] * 256 + image_safe[:, :, 2]
-
-            # You may want to cast the final result to uint16, but casting to a
-            # larger int type (like int32) is sufficient to fix the overflow.
-            # depth_map = depth_map.to(torch.uint16) # Uncomment if uint16 is strictly required
-            return depth_map.to(original_dtype)
+            # Return int32 — the 16-bit value does not fit back in the uint8 input dtype.
+            return depth_map
 
         elif isinstance(image, np.ndarray):
-            # NumPy equivalent: Cast to a safe dtype (e.g., np.int32)
-            original_dtype = image.dtype
             image_safe = image.astype(np.int32)
-
-            # Calculate the depth map
             depth_map = image_safe[:, :, 1] * 256 + image_safe[:, :, 2]
-
-            # depth_map = depth_map.astype(np.uint16) # Uncomment if uint16 is strictly required
-            return depth_map.astype(original_dtype)
+            # uint16 matches the mode="I;16" consumer in numpy_to_depth.
+            return depth_map.astype(np.uint16)
         else:
             raise TypeError("Input image must be a torch.Tensor or np.ndarray")