I tried out the new generative expand in all directions thingy today. At first glance it looked quite impressive. Closer inspection revealed a few flaws.
Problem 1: I drag out my canvas in PS using the crop tool - then expand the original image. The generated expansion doesn't fill the canvas - there's an annoying small gap around the outside.
Not sure if this is just my settings, Win 10 or a general problem.
If other people are affected can we get the generated fill to exceed the size of the canvas by a few pixels please.
Problem 2: The generated expansion alters the original considerably. Example. - picture of house has small tower - blue sky above - trees to the right - hedge to the left.
Expand doesn't give me more blue sky - it adds a crenallation to the top of the tower, what appears to be disgarded rubbish appears in the new hedge to the left, the trees to the right get taller and the conifer to the extreme right gets replaced completely by a deciduous tree.
Some of this (where it overlaps ) I can paint out on the mask - but the rest remains.
I know this is new technology but I would suggest that, going forward, generative expand needs to be fundementally different to generative fill.
I propose that expanding an image should give results which are more faithful to the original.
Generative fill is about imagining new items to add over an existing image. Generative expand needs to seamlessly expand the original without adding unnecessary new items.
If there is blue sky above we don't need bits added to make buildings taller, we don't need to make the trees taller and we most certainly don't need artificial additions such as random bits of trash and alien trees. Such additions should be 'fill' not 'expand'.