This is the Pony XL version of my former "enhancer" module, trained on more data.

It can work on a wide range of strengths all the way down to 0.3, but somewhere between 0.7 gives the best results.