diff --git a/changelogs/master/fixed/20250321_fix_cv_shape_error.md b/changelogs/master/fixed/20250321_fix_cv_shape_error.md
new file mode 100644
index 000000000..f01a4d237
--- /dev/null
+++ b/changelogs/master/fixed/20250321_fix_cv_shape_error.md
@@ -0,0 +1,12 @@
+# Fix error in `arithmetic._add_scalar_to_uint8_` on Windows
+
+On Python 3.8 tests for `_add_scalar_to_uint8_` fail for
+windows due to dimension and dtype-specific behavior
+in `cv2.add`.
+
+The fix the errors by directly adding scalar values
+if the input image is 2D or 3D with a singular channel
+dimension.
+
+If channel-dependent is required, broadcast the values
+to full image shape and add it via `cv2.add`.
\ No newline at end of file
diff --git a/imgaug/augmenters/arithmetic.py b/imgaug/augmenters/arithmetic.py
index 84f39bf9f..bbaee9d88 100644
--- a/imgaug/augmenters/arithmetic.py
+++ b/imgaug/augmenters/arithmetic.py
@@ -166,25 +166,18 @@ def _add_scalar_to_uint8_(image, value):
         value = round(value)
     elif ia.is_np_scalar(value) or ia.is_np_array(value):
         is_single_value = (value.size == 1)
-        value = np.round(value) if value.dtype.kind == "f" else value
+        value = np.round(value).astype(int) if value.dtype.kind == "f" else value
     else:
         is_single_value = False
-    is_channelwise = not is_single_value
-
-    if image.ndim == 2 and is_single_value:
-        return cv2.add(image, value, dst=image, dtype=cv2.CV_8U)
 
     input_shape = image.shape
-    image = image.reshape(-1, 1)
-    values = np.array(value)
-    if not is_channelwise:
-        values = np.broadcast_to(values, image.shape)
-    else:
-        values = np.tile(values, image.size // len(values))
 
-    image_add = cv2.add(image, values, dst=image, dtype=cv2.CV_8U)
+    if is_single_value and (image.ndim == 2 or (image.ndim == 3 and image.shape[-1] == 1)):
+        values = value       
+    else:
+        values = np.broadcast_to(value, input_shape)
 
-    return image_add.reshape(input_shape)
+    return cv2.add(image, values, dtype=cv2.CV_8U).reshape(input_shape)
 
 
 def _add_scalar_to_non_uint8(image, value):