I work a lot on stuff like these (for various reasons… DeDots - AleInno Design )
I see you are trimming the jpg image to be sampled, in such a way to have image areas that are somehow already on their own following the shape of your final panels: I do advice against that 
I always get much better results by allowing the points in GH to define the project area, and treat the whole image as canvas with extra content that can be sampled anywhere, like this:
which means to “untrim” the jpg image, in such a way there will be white_pixels_of_flowers also outside what you assumed would be the boundary of each panel (because there will most probably be)
the other thing I would note is that when you define your “basic grid” for Orienting the leaf pattern, if that basic grid does not already perfectly fit inthe boundary of your panels, then also the leaf pattern won’t
here, already at isotrim, you see that the grid is off alignment:
this means that you won’t get a good alignment from the very start
best way to do it: first in Rhino use the DIR command in such a way to make sure all the UVs of your surfaces flow in the same direction, from this:
to this:
then isotrim similarly to what you have already done, and then take care of replacing the image file with one that does not have panel trimming as explained above
and also consider that you can’t have scale at 0%, so this left point of the graph mapper should never go to Y=0
these vertical rows of tiny scaled leaves are the byproduct of using an image with divisions:
replace the image and those will be gone 
Magnolia Oxford Metal Screen Script for fourm_inno.gh (2.3 MB)
surfaces are internalized, no need to post 3dm file, but you might want to relink them to your own Rhino file eventually
Random example based on a different pattern:
