Memory errors in spite of adding capabilities (sometimes)

[RexLuscus]

In spite of adding the two required capabilities, I'm still (sometimes) crashing due to memory. Haven't figured out a consistent way to reproduce it, but it looks like a 50/50 chance to succeed, and once I have a failure in a session, generation continues to fail.
Model is Stable Diffusion 2.1, XCode is 14.3, device is iPhone 13 Pro, running iOS 16.4.1. Error in the log is imggensd2[1787:30752] [espresso] [Espresso::overflow_error].

Will be happy for ideas on what to check.

[ynagatomo]

Sorry. I have no ideas about this issue. So far, it seems the execution on iPhone is very difficult.

By starting over from recreating the Image generation pipeline, continuous failures may be improved somewhat. But it doesn't really solve the problem.

Anyone have any good ideas?

[RexLuscus]

I did follow someone's advice to try it in the simulator, and indeed in the simulator it seems to work consistently. Memory consumption is very high (around 2.7GB at the peak of the generation cycle). Strangely enough, there's no Jetsam event for the termination on iPhone, which I guess means the application is terminating itself? Anybody has good experience with running this on iPhone?