After running the script, you end up with separate puzzle pieces, which you can move around at will.
The script assumes a single object or group is selected before running. It asks for a few simple parameters, such as how many rows and columns should be created. It does not work in the faux perspective of your example, it assumes a plain orientation. If you want the effect you show, first transform the entire image in a proper "upright" position, then run my script, select all puzzle pieces, and transform them back to the slanted orientation.
Note that it creates a full copy of all of the objects for each separate piece. If your image is object-heavy (an autotraced image usually is), consider rasterizing it before running the script.