Question about incorporating image context persistence

[Kwisss]

Hi everyone,

I'm working on a chat GUI using llama-server.cpp and flask/js, and I'm running into an issue with maintaining image context across turns. Currently, the image from my opening message is reprocessed on each interaction, which is a bit slow. Or the model needs to rely on its initial response.

Any suggestions or insights on how to best manage image context (embedding?) within the current ecosystem would be greatly appreciated.