diff --git a/src/deepali/data/flow.py b/src/deepali/data/flow.py
index 75ea3b2..16f6d63 100644
--- a/src/deepali/data/flow.py
+++ b/src/deepali/data/flow.py
@@ -12,7 +12,7 @@
 from deepali.core.enum import PaddingMode, Sampling
 from deepali.core.grid import ALIGN_CORNERS, Axes, Grid, grid_transform_vectors
 from deepali.core.tensor import move_dim
-from deepali.core.typing import Array, Device, DType, EllipsisType, PathStr, Scalar
+from deepali.core.typing import Array, Device, DType, EllipsisType, PathStr, Scalar, ScalarOrTuple
 
 from .image import Image, ImageBatch
 
@@ -203,12 +203,25 @@ def axes(self: TFlowFields, axes: Optional[Axes] = None) -> Union[Axes, TFlowFie
         data = move_dim(data, -1, 1)
         return self._make_instance(data, self._grid, axes)
 
-    def curl(self: TFlowFields, mode: str = "central") -> ImageBatch:
+    def curl(
+        self: TFlowFields,
+        mode: Optional[str] = None,
+        sigma: Optional[float] = None,
+        spacing: Optional[Union[Scalar, Array]] = None,
+        stride: Optional[ScalarOrTuple[int]] = None,
+    ) -> ImageBatch:
         if self.ndim not in (2, 3):
-            raise RuntimeError("Cannot compute curl of {self.ndim}-dimensional flow field")
-        spacing = self.spacing()
-        data = self.tensor()
-        data = U.curl(data, spacing=spacing, mode=mode)
+            raise RuntimeError(f"Cannot compute curl of {self.ndim}-dimensional flow field")
+        if spacing is None:
+            if self.axes() is Axes.GRID:
+                spacing = 1
+            elif self.axes() is Axes.WORLD:
+                spacing = self.spacing()
+            elif self.axes() is Axes.CUBE:
+                spacing = tuple(2 / n for n in self.grid().size())
+            else:
+                spacing = tuple(2 / (n - 1) for n in self.grid().size())
+        data = U.curl(self.tensor(), mode=mode, sigma=sigma, spacing=spacing, stride=stride)
         return ImageBatch(data, self._grid)
 
     def exp(
@@ -529,10 +542,16 @@ def write(self, path: PathStr, axes: Optional[Axes] = None, compress: bool = Tru
         disp = disp.axes(axes or Axes.WORLD)
         Image.write(disp, path, compress=compress)
 
-    def curl(self: TFlowField, mode: str = "central") -> Image:
+    def curl(
+        self: TFlowField,
+        mode: Optional[str] = None,
+        sigma: Optional[float] = None,
+        spacing: Optional[Union[Scalar, Array]] = None,
+        stride: Optional[ScalarOrTuple[int]] = None,
+    ) -> Image:
         r"""Compute curl of vector field."""
         batch = self.batch()
-        rotvec = batch.curl(mode=mode)
+        rotvec = batch.curl(mode=mode, sigma=sigma, spacing=spacing, stride=stride)
         return rotvec[0]
 
     def exp(