The first model is nice and I will keep it, but it's very hard fit all kinds of images with just one model. The new model is worse in some points, but it doesn't have the only big problem the older one has: great lost of mid-tone information.
This new model usually is more useful for photos with at least 1 of 2 properties:
- high detail level;
- low contrast.
If the input photo has both features, so probably this new effect can be even more useful.
The example shows you a canonical use of this new model: it tries to keep most of original photo details (we miss this details even more since most of them are geometric). The older model is more stylistic, but can not cartoonize this photo keeping a high detail level.