Content Aware Fill v2

Could we have an updated version of Content-Aware Fill, please?
The new one would utilize more than one reference frame. It would also be more precise even at the cost of calculation speed.
The current version is rarely useful due to poor content generation.
Thanks.