Samples show good results, however....
I tossed 4 projects which I am currently working on at it, and the process, in all the different set-ups I gave it, failed miserably. Completely unacceptable results. On top of that, the interface is gawd-awful.
I gave it a 1x2 and 2x2 photographic and a 1x3 and 2x2 multipage scan.
The 2x2 photo of a steppe-villatge was hilarious, as it put 3 of the shots exactly on top of each other and fish-eyed them, and put the lower-right image immediately to the right of the stack. Awful.
The 2x2 multipage scan turned into something resembling a doily (or mirrored cylinder art), arranged 1x4.
And the very simple 2 image photo with 1/3 overlap, it couldn't even put together, just saying it couldn't find a panorama.
This has got a very long way to go for anything I could actually use. In the time it took to fail with the two shot photo, I could have stitched it in photoshop, with only the central distortion requiring a correction.
I could use a tool to do this with my hundreds of multi-page scans that I deal with on many projects, but it actually has to work... And my scans don't have typical lens, focus, and dof distortions... These should be easy to deal with. (And how about a switch turning OFF all features that require JFIF information, as not everybody just plugs their camera in and drops the photos immediately into a stitcher.)
Some people might get results with this, and other reviews seems to support this, but I'm certainly not getting good results, and I believe that allowing the user to pre-set-up relative positions, sizes, and co-incidence points would give the program a better chance to select proper stiitch points.