The solution I have used for panoramas that are difficult to stitch does not use move images.

The method that takes the longest but is the most reliable for me goes like this:

1) I put all of the images into a detect so as to find two or more images that are linked together as one panorama through automatic detection then delete all the other images: this is the core to which I am going to add images in turn

2) I have the candidate images displayed in another browsing application, I use Picasa. This allows me to pick out images that I know are neighbours of each other; I have a standard way of taking panoramas that helps with this sequencing.

3) I use the Add Images feature of the Panorama Editor to add in a next unlinked image that I know is a neighbour of the sub-set created at step 1)

4) Switching to the Control Points Editor(CPE) the added image will be the last in the image list, and shown in red as having no links. I select in the image list what I know is a neighbour to the new image and the new image itself, draw marquees in a selected area common to both images and choose 'Add Control Points' - if CP are not found it can help to change the CP detector in Settings to 'high'

5) When new control points have been added I then run optimisation, in the CPE

6) I return to step 3) until all the candidate images have been added.

Two comments

a) Yes, this technique is not recommended for very large panoramas; but I have used it on panos with 25-30 images

b) if no CP are found at step 4) then I look for another known neighbour and try again: if no known neighbours will link in then I am sunk - but that has not happened to me, as yet

leedsjoe wrote:

if no known neighbours will link in then I am sunk

If I understand your technique, then this is exactly the problem I am interested in. Eg typically I'll have some adjacent tiles of sky that are have some cloud structure visible, but nothing that the Add Control Points tool can use.

A

DrSlony wrote:

1- choose the move hand tool, hover over that stack and use your mouse wheel .

The mouse wheel? Is this documented? It certainly works - thanks!

Aeriscira: apologies for not reading your original post carefully enough. On re-reading, I see you had tried 'add control points' before trying the Move Image Feature. (This said, using add images to bring in unlinked images one-by-one rather than using forcing to add all unlinked images in at the same time may make for an easier workflow.)

Some matters arising which may help generally - I have just checked that these points also apply in the version of 1.9 released today.

1) about the user adding/inserting their own control points by hand: this has been often requested but is still, as far as I know, not implemented

2) the use of the mouse wheel is documented in the wiki under geometry editing within layers mode: it also appears in the help text generated when clicking on the help icon, the question mark, in the Panorama Editor, then on the layers mode icon

3) the hard link feature is a right-click on the image number in the Panorama Editor rather than the CPE but the CPE has to be 'on' in order to access hard link. Hard Link is not available with Layers Mode on, or directly in the CPE. (At this point, I feel like repeating the old request not to shoot the messenger.)

I have used the search in the wiki to see if I can find 'hard link' documented, without success, so I am not clear what benefits you get from using it - perhaps others could help out here.

Lastly, a perhaps crazy thought about handling images with very few features in them, e.g. sky with ill-defined clouds. One could use an add text feature in an image processor to locate identical markers in what you know are the same spot in the sky next to the 'same' clouds. Then APP could stitch using those markers and the user could, later, clone them out. Ok, a desperate work-around but has anyone used it successfully?

This is one of the areas where shooting a giga pano with a robot is an advantage because the yaw and pitch are more consistent than when shooting it manually.  I shoot a different pattern when I shoot manually and my pitch is very consistent, but my yaw has much more variance.  This has solved my problem with stitching sky which may have beautiful cloud and light patterns, but not enough details to be automatically positioned correctly.

1.  I choose to include all images in the panorama before I tell it to detect the panorama.

2.  When I shoot using a robot and I use AP Giga, I am able to edit the incorrect yaw and pitch values for pieces of the sky by looking at the yaw and pitch values for adjacent photographs.  I do not move the incorrectly positioned photos using the mouse, but I double click on the table's yaw and pitch values and enter a better value.  Then it does NOT need to reproject the entire panorama.  It only modifies the area where the photograph has moved to. (KEY for very large panoramic images)

3.  Then I tell AP Giga to optimize the color again and it looks great.