Hugin output doesn't match preview

Trying to create a panorama in Hugin required moving the control points (the red X’s ) from the clouds to the ground and manually aligning the images so that the horizon is continuous. (I’ll go back and blend the clouds in GIMP.)

Here’s how it looks in Fast Preview (which I know is “less accurate” than the Preview, but every time I use the Preview window it does weird things):

Then stitching the images results in an output different from the GUI - perfect clouds and a broken ground.

Any way to fix this?

Choose more points manually and make sure you have the lens correction dialed in. Your fast preview has obvious seams where the output file also has issues.

Not sure sure how you do lens correction effectively. Every time I tried to optimize the lens:
(Interface>Expert>“Photo” Tab>[Expert Interface] Group By…> Lens>Optimize Now!),

the panorama would only get wackier. Even adding 40 control points manually on each image:
(Layout>Click photo>Click point on image… automatically finds mate>right-click to accept)

didn’t affect the end result. Neither did autogenerating CPs in the Feature Matching section under photos, clicking “Fine Tune” when adding control points, or changing the “Mode” of a control point to “horizontal line”.

As you can see, it’s nearly flawless in the preview, yet the output never changes. It’s almost like the preview changes are being recognized by the rest of the program.

Preview:

Export:

May have to crop out the sky on all the pictures, and then blend the ground and sky separately and fuse them at the horizon.

You have control point near the camera and far away. If you did not rotate the camera about the non nodal point you will have parallax issue - which is here probably the culprit.
Remove the control points near the camera.
But just adding or removing control points is not enough. You need to reoptimize the pano afterwards - then the new or removed control points will be taken into account and will have effect.

When all control points are in a very narrow area of the images you should not optimize the lens distortion parameters. There is then too less spreading of the control points to optimize the lens distortion parameters. Try only optimizing image positions (and maybe field of view).