New to version 1.6, Photosphere can now stitch together some pretty nice panoramas. It's a little awkward to use, but it works. (See below for some tips on how to go about it.)
Here is an example output from Photoshop CS2, showing typical problems from joining hand-held panorama inputs:
Here is the same set given to Photosphere:
Here is a nice sunset -- click on the image to see it at half resolution, and download the link for the JPEG-HDR version:
As promised, here are some tips on how to use this new feature in Photosphere, quoted from quickstart_pf.txt:
15) To stitch together a panorama, choose a set of images and open them. In each image, select a salient feature that is also visible in the adjacent image. Click the "Apply" button in the image viewer and choose "Pano Orig" in the first image, then move to the second image, select the matching feature, and choose "Pano Dest". After a moment, the selection box in the second image should disappear, indicating these two images have been linked. If the "Apply" button is missing "Pano Orig" or "Pano Dest" in its pop-up menu, that means the box is either too big or in the wrong image. If you have previously selected a set of anchors or created a panorama and wish to start a new one, use the "Clear Pano" command that shows up in the "Apply" pop-up when nothing is selected. After each link has been made, you may select the next feature for linking to another neighbor, or choose "Make Panorama..." from the "File" menu if you are done or wish to see what you have linked so far. By scaling down the output resolution, you can obtain a quick preview to check your work. You can always add new linked images or change existing links via the "Apply" pop-up menu, or start over by selecting "Clear Pano" or pressing the "Clear" button in the "Make Panorama" dialog. It may take a few tries before you get used to the idea of selecting image features, and some features definitely work better than others. It's best to select a corner or prominence with high contrast. Busy areas make for poor feature matching, which can cause misalignment in the results. Trees and mountain peaks are good features. Fields of grass and waves on water are not.