diff --git a/fastplotlib/utils/functions.py b/fastplotlib/utils/functions.py
index 97a3df742..9b6c83c6c 100644
--- a/fastplotlib/utils/functions.py
+++ b/fastplotlib/utils/functions.py
@@ -408,14 +408,9 @@ def parse_cmap_values(
 
 
 def cuda_to_numpy(arr: CudaArrayProtocol) -> np.ndarray:
-    try:
-        import cupy
-    except ImportError:
-        raise ImportError(
-            "`cupy` is required to work with GPU arrays\npip install cupy"
-        )
-
-    return cupy.asnumpy(arr)
+
+    data = np.from_dlpack(arr, device='cpu')
+    return data
 
 
 def subsample_array(
diff --git a/fastplotlib/widgets/nd_widget/_base.py b/fastplotlib/widgets/nd_widget/_base.py
index 5ca15889d..ce17baef7 100644
--- a/fastplotlib/widgets/nd_widget/_base.py
+++ b/fastplotlib/widgets/nd_widget/_base.py
@@ -198,6 +198,13 @@ def spatial_dims(self, sdims: Sequence[str]):
 
         self._spatial_dims = tuple(sdims)
 
+    @property
+    def spatial_dims_indices(self) -> tuple[int, ...]:
+        """
+        The ordered spatial dim indices that correspond to the named spatial dims
+        """
+        return tuple(self.spatial_dims.index(d) for d in self.dims if d in self.spatial_dims)
+
     @property
     def tooltip(self) -> bool:
         """
@@ -529,12 +536,7 @@ async def get_window_output(self, indices: dict[str, Any]) -> ArrayProtocol:
                 f"windowed_slice.ndim != len(self.spatial_dims): {windowed_slice.ndim} != {len(self.spatial_dims)}"
             )
 
-        # transpose to spatial dims
-        spatial_dims_int = tuple(
-            self.spatial_dims.index(d) for d in self.dims if d in self.spatial_dims
-        )
-
-        return windowed_slice.transpose(*spatial_dims_int)
+        return windowed_slice
 
     async def _get_raw_data_slice(self, indices: dict[str, Any]) -> ArrayProtocol:
         """
diff --git a/fastplotlib/widgets/nd_widget/_nd_image.py b/fastplotlib/widgets/nd_widget/_nd_image.py
index 3090e14c7..40dd510f9 100644
--- a/fastplotlib/widgets/nd_widget/_nd_image.py
+++ b/fastplotlib/widgets/nd_widget/_nd_image.py
@@ -258,7 +258,7 @@ async def get(self, indices: dict[str, Any]) -> ArrayProtocol:
         if isinstance(window_output, CudaArrayProtocol):
             window_output = await run_in_thread_pool(self._executor, cuda_to_numpy, window_output)
 
-        return window_output
+        return window_output.transpose(*self.spatial_dims_indices)
 
     def _recompute_histogram(self):
         """
diff --git a/fastplotlib/widgets/nd_widget/_nd_positions/_nd_positions.py b/fastplotlib/widgets/nd_widget/_nd_positions/_nd_positions.py
index 2cc768f52..9ca5eee93 100644
--- a/fastplotlib/widgets/nd_widget/_nd_positions/_nd_positions.py
+++ b/fastplotlib/widgets/nd_widget/_nd_positions/_nd_positions.py
@@ -112,6 +112,7 @@ def __init__(
         self.cmap_transform_each = cmap_transform_each
         self.sizes = sizes
 
+
     def _check_shape_feature(
         self, prop: str, check_shape: tuple[int, int]
     ) -> tuple[int, int]:
@@ -555,6 +556,8 @@ async def get(self, indices: dict[str, Any]) -> dict[str, ArrayProtocol]:
         if isinstance(data, CudaArrayProtocol):
             data = await run_in_thread_pool(self._executor, cuda_to_numpy, data)
 
+        data = data.transpose(*self.spatial_dims_indices)
+
         return {
             "data": data,
             **other,
diff --git a/fastplotlib/widgets/nd_widget/_nd_vectors.py b/fastplotlib/widgets/nd_widget/_nd_vectors.py
index 138ddee95..1a4d1b8e5 100644
--- a/fastplotlib/widgets/nd_widget/_nd_vectors.py
+++ b/fastplotlib/widgets/nd_widget/_nd_vectors.py
@@ -157,7 +157,7 @@ async def get(self, indices: dict[str, Any]) -> ArrayProtocol:
             Example: get((100, 5))
 
         """
-        # this will be squeezed output, with dims in the order of the user set spatial dims
+        # this will be squeezed output, with dims in the order of self.dims
         window_output = await self.get_window_output(indices)
 
         # apply spatial_func; CUDA arrays run inline, numpy goes through the thread pool
@@ -175,7 +175,7 @@ async def get(self, indices: dict[str, Any]) -> ArrayProtocol:
         if isinstance(window_output, CudaArrayProtocol):
             window_output = await run_in_thread_pool(self._executor, cuda_to_numpy, window_output)
 
-        return window_output
+        return window_output.transpose(*self.spatial_dims_indices)
 
 
 class NDVectors(NDGraphic):
diff --git a/pyproject.toml b/pyproject.toml
index 0352cf27c..9c914bd79 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -19,7 +19,7 @@ keywords = [
 ]
 requires-python = ">= 3.10"
 dependencies = [
-    "numpy>=1.23.0",
+    "numpy>=2.1.0",
     "pygfx==0.16.0",
     "wgpu",          # Let pygfx constrain the wgpu version
     "cmap>=0.1.3",