you can inject context to your bot using the Rasa HTTP API by appending SlotSet event to the tracker.
Assuming here you have some kind of web front end(react, vue etc), you can take the image context, and create a slot in Rasa domain, create stories with the context of this slot value and add value in real time to the slot using the SlotSet Event, this way, the bot already has the context of what the image is about, thus any question triggered on this image page, would be about the image already.