you can have images and they don't always display as anything. the author of the user agent decides how they are displayed. you could in-line them if you wanted to, but only a few clients do that at the moment. there are no hints to the client about how some content could, should, or should "always display as".

It's text. The client displays that text and renders links, headings, etc, however it wishes. If it really wants to, it could just not format them at all. There's a gemini client made for plan9's acme text editor that doesn't render links, and instead displays them verbatim, because the plan9 plumber can handle the hyperlinking aspect. All of that is eye candy and fluff.

If a client finds a link to an image, it can in-line it if it wants. If you wrote a client, when it found a link to an image, it would in-line it "with nothing to either side." That's not something that has to be specced.

You could have a setting on the client that lets the user specify that that's how they always want to see images, I on the other hand might specify that I want to open them in a new window, or see a thumbnail until i hover over, etc.

Same with how headers are displayed (maybe i want folding or something), whether an ToC is displayed, colours, fonts, etc.

The point is that the user can decide all this stuff, without having to hack it around the author's own styles and scripts.