Here’s what is happening, as far as I can tell: Since the image you are bringing in is larger than the dimensions of the Canvas, Motion is scaling the image to fit inside of it. When you draw the shape, Motion is using the dimensions of the Canvas as the reference for the attributes of that shape. Thus, when you try to drag it onto the image to create a mask, Motion is still using that information to determine where on the image relative to the canvas it should create the mask. Since the image is bigger, the position is going to inevitably be off.
There are a few ways to get around this:
1. Use the Bezier Mask tool directly on the image rather than drawing a shape and then converting (by dragging, as you did). You shouldn’t have any issues doing it that way.
2. Draw the shape like you did and use it as an image mask on the image, rather than a regular mask.
3. In Motion’s Preferences, under Project there is a dropdown at the bottom that will say Large Stills: change it to Do Nothing. When importing your image it will go beyond the bounds of the canvas, but the shape-to-mask will work as you are expecting it to. You can then either scale the image or use a camera to get the layout you want.