Well I have been wanting to answer your question ever since you posted it, but I don't know if my comments will be useful to you. Most of the descriptions I encounter and consume are in the form of AD tracks on movies (and
$1 $1 streaming TV), not alt-text. For all I know the rules of the two formats are completely different, and the expansive descriptions I prefer in movies would not be suitable for an alt-text. But I figure that basically, the foundations and standards of good description writing are the same as any other kind (we don't need specialised prose), so here are a few thoughts, for whatever they are worth:
* I once heard from a blind commenter on here that it is considered bad form for alt-text to exceed two or three sentences; when I Google "alt-text how long", the results tell me that an optimal length is between 80-125 characters. From personal experience this feels like an underestimate for high-quality, human-written descriptions like the ones you are trying to make, which are double that length or more. This is pretty much your only textual limitation, and for the purposes of an artistic image as opposed to an informational one, I think you should opt for descriptions on the longer, richer side if you want to give us a taste of what we are missing. Generally, more description is better, although with alt-text it becomes a balancing act where you're trying to convey the maximum amount of information in a minimal number of words. The AD tracks that have the greatest impact on me (the ones that can make you laugh, cry, and feel happy at the same times that sighted people do) have these two things in common: 1) the describer never shuts up they tell you every-thing, and…
* 2) their prose has personality and spark, a certain sense that a real, engaged human being wrote the words we are listening to instead of a robot or a bored group of writers that just wants to hammer the shit out and go the hell home. Good descriptions can even pique your interest in something you would normally skip over, like the stylised titles at the beginnings of films. From Ingrid goes West, as described by one of the greatest AD writers currently working, who has an unspellable name that sounds something like "Tae Se": "Against the image of a perfect empty beach with azure waves gently lapping at the golden sand under a deep blue sky, *Ingrid Goes West* appears in flickering pink neon lettering." [180 characters, all of her description is written like this, she is a one-woman DVS]. You don't always have to be poetic and creative either, evocative descriptions can be made from short, clear, functional sentences as well, if you know what you are doing. (From Greener Grass: Jill, whose son Julian has just been transformed into a dog at a party, spots a stranger watching the whole ordeal. The stalker gets herself out of sight, and over the sound of panting, the describer distractedly notes that "Dog-Julian continues to sit like a goodboy." [possibly done by Audioeyes, uncredited scriptor and voice.]) For that reason, I would be inclined not to impose any grammatical suggestions or style guides on you, because it is precisely your own voice, whatever it might sound like, that enhances the descriptions and makes them
$1* Technical: The two screenreaders I use split webpages into around 150 characters per line; if your description is longer than this nothing will be cut off, we just get an alt-text with two or more lines. Putting in your own custom line breaks is a little dodgier, you don't know if Instagram will respect them. I've seen alt-text where it looked like the writer had tried to split their description/joke into multiple lines (imaginary dialogue, quoted poetry), but the website mashed it all into one continuous block. The safest thing to do, guaranteed to work every time, is to format your text as if it were a paragraph from a novel or essay. Shorter descriptions followed by an optional expansion would be immensely useful so that we could choose the approach we want, but I've never been on Instagram and have no idea how it would work, you might end up needing to go with one style.
* Finally, the best way to learn how to do this is simply to read text-based descriptions yourself as you beef up your skills. This is also not-about-alt-text and you will need to skip the stuff about the layout of strips and panels, but it's a really useful resource if you're looking for lingual inspiration and you want to see how an amateur pro does it: Liana Kerr describes comic books for her blind friend, and made some of them available on the web for anyone to read.
$1 are her philosophies,
$1 is a huge sample, a complete descriptive transcript of Watchmen.