I prefer the lack of inline display; I would rather it not display the pictures inline in the document anyways. However, that is something that the client does anyways, rather than the server. (Also, I think it does allow inline display if the picture and the document linking to it are both local files (or are both inside of a ZIP archive which is a local file), although even then it should have the option for the user to disable them.)
(Furthermore, it seem to me that Gemini is not a very good protocol for serving a video file anyways, because it does not have Range requests and stuff like that.)