It's twofold: 1. (x)diffs of binary files are usually more expensive than text files. For instance, consider a git-tracked bitmap image where you apply a glob

It's twofold: 1. (x)diffs of binary files are usually more expensive than text files. For instance, consider a git-tracked bitmap image where you apply a global filter. If you add the revised image, and every pixel was touched, this will usually result in a large diff. So, git will either store the large diff, or the modified file completely. If you, on the other hand use SVG with filter effects, the application of a filter will just add a few lines of text to an SVG file. This gives a small diff, meaning that the object can be stored compactly. 2. In the SVG scenario, you could easily spot the changes between both revisions of the image with a 'git diff' (since both files are plain-text). In the bitmap scenario, you can only spot the differences visually.