List of steps to reproduce:

• Edit any image using VE
• Add caption data in the VE Media dialog
• Change image format from a "block" format (thumb or frame) to an "inline" format (frameless or border).

What happens?:
Changing from a block format to an inline format loses caption data.

What should have happened instead?:
Caption data should persist with the image, even if it is not rendered because a different format was chosen.

Software version (if not a Wikimedia wiki), browser information, screenshots, other information, etc:
Found in REL1_35, but seen in the modules/ve-mw/ui/dialogs/ve.ui.MWMediaDialog.js code for master.

If I understand it correctly, the current code looks for a DOM element which is not there if the image format renders in inline mode? For inline modes, MediaWiki could create a hidden DOM element to preserve caption data.