This ticket identified a bug that made it impossible to edit captions for many files on Beta Commons. The problem was fixed for files with existing structured data as of this patch and for those without any existing data with this patch.
Ideally, the WikibaseMediaInfo extension could rely on the wikibase.entityPage.entityLoaded hook to for initialization. However, this hook currently fires only if structured data already exists for a given file. Newly-created files lack this data, and therefore the captions panel UI was broken on these pages prior to the recent workaround.
It seems like there are two possible solutions here: either 1) ensure that the wikibase.entityPage.entityLoaded always fires (a Wikibase change), or 2) changing the way that CaptionsPanel gets initialized (a MediaInfo change).
If we go down the latter route, some refactoring may be necessary in the CaptionsPanel class to allow for a widget to be created with no data and then (optionally) have that data provided at a later time based on whether or not the relevant hook gets called. This class is one of the most complex bits of JS in the extension so it may be a non-trivial change.
Let's use this ticket to discuss how to move forward.