I wish something like this on gmic
Interesting. I'm really surprised that just 68 image pairs (plus some fuzzing) were enough to train that model.
68 image pairs make tens of thousands individual strokes...
As this uses a neural network I think this post from @David_Tschumperle is the answer:
This is a bit simpler.
potrace no this so far of the result
Using Engrave, here's the results I got.
Why not the original target ones too.
Those have a lot of double-lining that the Sketch Simplification algorithm is supposed to eliminate.
Maybe so, but a close approximation and with slider tweaking, could get even better results (didn't really spend that much time playing with the sliders).
I don't think you need neural networks just to simplify double lining. I'm a bit surprised, but the engrave filter here does a very good job. Adding some post-processing to the engrave filter should do the job as well.
This has helped me with an animation no very clean.