>>And, ah, the loFrm.img.picture=lcFullp still works, but... in many cases I'm getting fewer bitplanes - the rendering of the image is rather ugly and looks like it's reverting to 16 or 64 or 256 colors. So loadpicture()... which works.
>
>I wonder if any dangling references are piling up? Maybe what you were doing was a code path that wasn't hitting normal/expected cleanup code.

I was wondering the same... the iPicture object being passed back from a subroutine and then used to paint the form with a bitmap, then discarded (loobj=null) used to increase the memory consumption by random amount, usually just 4K (which is weird, as I'm ignoring images smaller than 8K). Looking at the task manager, it would jump by 10-15M (which looks about right, 6 images of about 1-2M being processed each time), then fall back to what it was, except it would add those 4K (most of the time, sometimes more). In the log, before the crashes, I've found more of the .png files, which are sort of suspect. During early testing I've found png files from some sources to be repeatedly causing OLE errors with the iPicture object. Perhaps when these errors happen, then it remains dangling? The process of finding out would take more time than I can allocate for a hobby project.

>I think source is available for all of VFPX (?) If so you could check it out, contribute reliability patch(es) :)

The code is in the help file only...