I hold my pencil betwix my paws

Type of media. Always assume that the reader is someone who has no idea who you are, or what your blog is, or what the context is. Is it a drawing? A meme? A screenshot? A purposefully bad photoshop? Etc. This is especially important if you draw something that would be very concerning if it wasn't a drawing. "ID: a corpse" is fucking crazy, "ID: drawing of my OC lying on the ground with Xs for eyes and a cartoon ghost leaving their body" is cuties.

If there's a specific art style happening, you can include it. Realistic, semi-realistic, anime, cartoon, surrealist, abstract... If it matters to the concept (ex. "tried a different artstyle for this one!") or is something that would be actively noticed by a sighted person (ex. redraw of some very serious subject but it's the damn family guy artstyle) you should include it.

Vague one line about the subject of the drawing to give a general concept of what the person should be imagining. You can add the relevant details later. "ID: a X is doing an Y. End ID." type of deal.

IMO this is what a bare minimum, but very much effective art description should be. This is the stage where you get a "ID: painting of a girl taking a picture of a colorful bird outside her window" type of description. It probably doesn't convey everything, but it gives the person a reasonable idea of what they're dealing with. This amount of alt text is always better than no alt text. I think it's also important to keep in mind that a lot of us that post art post it with objectively nonsense captions, so if someone has the post text of "Everyone look at this Now ❗" but doesn't have anything for the image they're supposed to be looking at, they won't know if it's a blorbo situation or a screenshot of a report of the newest diarrhea outbreak.

Less bare minimum:

Describe the subject! For me it's usually a person so that's what I'll be using. You don't have to describe every feature (and shouldn't), but you can try to include the demographic, some unique feature, how they're dressed; things that a person would visually see. Do not be vague to the point of being confusing. If every sighted person looks at the painting and sees a blonde goth white girl, you can just go with that and not "feminine-presenting individual with pale skin and straight light hair, wearing black clothes". (If something is actually ambiguous then sure, but don't actively deprive someone of a concise description when every sighted person can see what's going on).

What is the subject doing? How are they doing it? Is she taking the picture nervously with a huge camera, or with her phone and an expression like she's collecting evidence on that damn bird?

Any sort of emotions happening. Is the entire scene happy, sad, etc.? I'm autistic and do not recognize emotions by facial expressions (when I draw them I largely just copy without understanding) so I'll write things out literally. "She's excited", "she has a big smile on her face".

After that, include things that you'd actually consider meaningful. What's on the bookshelf probably doesn't matter. The color of the bedsheets doesn't matter, unless it's a colorful bird print that adds some context to the whole picture. The cleanliness of the room doesn't matter, at least not enough to describe it beyond "messy/clean/cluttered".

I really like this part in the Against Access essay about blind accessibility:

" Giving quick reads without getting bogged down in details is an important skill for a spy doing live reconnaissance. But ASL interpreters are at first inhibited by notions of neutrality and objectivity. They start by offering something like “Walking by over there is a tall, thin, light-skinned person with curly dark hair down to here, wearing a white tank top, blue jeans, and brown boots…” They’re pleased that they’ve avoided race and gender. “No, no, no.” I brush my hands back and forth across their arm in vigorous negation. “That’s not the way to do it. The very same description could be applied to a gorgeous Latina in chic, expensive boots, who oozes money, or to a pasty, rangy white man, hair a mess, boots falling apart, maybe looking angry. Like, they’re the opposite of each other? Yet they share the same sanitized description.” Because of their fear of bias, I discuss four implicit safety nets to help them feel better about uttering an assessment: First, we’re not so fragile that saying something wrong will topple us. We know what we are doing. We—not they—are in charge of our missions. Responsibility lies with us, not with them.

[...] I cannot count the times I’ve stopped reading a video transcript because it started with a dense word picture. Even if a description is short and well done, I often wish there were no description at all. Get to the point, already! How ironic that striving after access can actually create a barrier. When I pointed this out during one of my seminars, a participant made us all laugh by doing a parody: “Mary is wearing a green, blue, and red striped shirt; every fourth stripe also has a purple dot the size of a pea in it, and there are forty-seven stripes—” “You’re killing me,” I said. “I can’t take any more of that!” Now serious, she said it was clear to her that none of that stuff about Mary’s clothes mattered, at least if her clothes weren’t the point. What mattered most about the image was that Mary was holding her diploma and smiling. “But,” she wondered, “do I say, Mary has a huge smile on her face as she shows her diploma or Mary has an exuberant smile or showing her teeth in a smile and her eyes are crinkled at the edges?” It’s simple. Mary has a huge smile on her face is the best one. It’s the don’t-second-guess-yourself option. My thinking around this issue is enriched by the philosopher Brian Massumi’s concept of “esqueness.” He exemplifies it by discussing a kid who plays a tiger: "One look at a tiger, however fleeting and incomplete, whether it be in the zoo or in a book or in a film or video, and presto! the child is tigerized… The perception itself is a vital gesture. The child immediately sets about, not imitating the tiger’s substantial form as he saw it, but rather giving it life—giving it more life. The child plays the tiger in situations in which the child has never seen a tiger. More than that, it plays the tiger in situations no tiger has ever seen, in which no earthly tiger has ever set paw." Just as the child and an actual tiger are not one bit alike, the words Mary has a huge smile on her face have nothing in common with the picture of Mary holding her diploma. Yet the tiger announces something to the world, its essence, and a kid can become tiger-ized and be tiger-esque, their every act shouting, I am a tiger. The picture of Mary at her graduation is shouting something, and the words Mary has a huge smile on her face are also shouting something. It is at the level beyond each actuality, in the swirl that each stirs up, that the two meet. "

Don't make it way too long, the same way a sighted person is probably not going to look at it for 5 straight minutes a blind one is not going to listen to a 5 minute long description. Don't describe every detail. The color of the shirt probably doesn't matter. Etc.

To kinda go with that excerpt above, I do cheat with my tutorial things (and probably also would with comics). In the visual reality I use arrows and other nonsense to connect different parts to each other. But I don't think there's a point to describing all that. So I just say something is captioned with XYZ, or is below that, not that it's on the opposite side of the canvas with a long ass arrow because it's devoid of actual information for someone experiencing it in text form.

For pieces that include a lot of text I also have some tips:

No all caps on normal words because they can be confused for acronyms by some screen readers. Use an exclamation mark if it's about loudness, skip if meaningless.

99% of the time it doesn't matter when something is underlined or italicized, or in a different font, or whatever else you can do with text. Mention if it's genuinely important or is part of a joke, etc.

Use PascalCase or camelCase when writing multiple words together, so that screenreaders read them properly.

Rarely you might need to make some changes for readability. For example in my last post I talked about MOPDII. In alt text I went with MOPD 2. Both are accurate and used, but one is ambiguous (ii or 2? It's an acronym, not a year or a sequel afterall) while the other one isn't. You might do the same for one billion punctuation marks or emojis, if the number is important just say that XYZ is followed by a comical amount of interrobangs or something.

Last thing: if posting the ID in the body of the post, don't do any formatting, weird colors, putting behind the read-more, etc. it all defeats the purpose of accessibility. I see people put the only ID as this tiniest possible font, often italicized, and I'm sorry to say but 1) it just looks ridiculous, 2) not everyone using IDs uses a screenreader.