This seems like you're putting in serious thought and care on this issue. I'll give you my take, with the caveat that I'm not speaking for everyone in the community.

Often when people write alt text they have a hyperfocus on objects and their spatial relationships. Your example with the broom is pretty representative of this. What's better is to go up one level of abstraction and think more about the purpose of why you're sharing or exhibiting the photograph.

Less useful: A photo of a smiling older woman wearing a pointy hat sitting at a table. There are candles on the cake, which are lit. There is a cat sitting on a windowsill in the background.
More useful: My grandmother's birthday party. She's about to blow out the cake, and we're all smiling.

If there is a person in a blue dress in your photo, you don't need to say the dress is blue unless it's salient to why you're sharing the photo or unless you anticipate a specific use case for the photo that requires that information.

If you want blind people to be able to participate in your exhibit—appreciated—I'd recommend going the narrative text route that you laid out. Articulate the meaning, purpose, and importance of what you see, rather than the minutiae. I wouldn't write reams of text—a picture might be worth a thousand words, but a blind person isn't going to read a thousand wrds for every photo. Fundamentally, a photo is a sighted medium, so there's some extent to which you'll be leaving blind people out by the exhibit. That's largely OK, your responsibility (moral if not legal) is to provide reasonable accommodations, and that mean the experience isn't 100% translated into an accessible format.

Hope that helps, happy to follow up if you ask questions.

Thank you for the clear levels of dedication and consideration you are putting into this project. There are some best practices around a visual description, which can serve as a good starting point. Discussing a photographs contents really depends on the photo itself. Different visitation patterns, like clockwise or quadrant-based or linear, can be used to help the person who can’t see the photo to build a mental model in which you can fill in all of the details you are speaking about. I’ve done a considerable amount of work in the space, and we regularly hold visual description workshops for people in the culture sector, especially museums. Happy to share specific tips and resources. Does your project allow for tactile reproductions? When coupled with guided tactile descriptions and visual descriptions, they can be a wonderful multi modal way of conveying visual information. This is the approach we took with the Andy Warhol Museum and the Canadian Museum For Human Rights, the latter of which actually has had multiple exhibits, traveling ones, consisting of lots of photographs. We’ve also found that making sure the description is able to be accessed both as a recording as well as programmatically, helps a whole lot. Someone like myself who prefers very rapid speech can easily read a 350 word description in 15 to 20 seconds, but someone else may prefer to listen to that 350 word description over the course of two minutes as a pleasant voice Recording. At the end of the day, this is about returning agency to the visitor, and there are some inexpensive ways of doing that as well as higher quality ways if you’ve got some budget for things like reproductions. Art Beyond Sight has some guidelines that we like to start with Wynn showing people how to do visual descriptions. There is good advice in those like how to use metaphor appropriately when discussing differences in colors.
Happy to talk more about this. Feel free to reach out.

I agree that the point of the image is the most important part. I've used many analogies to explain it in the past, but a logo is an easy way to show what I, personally, like.

From real bad to good:

- logo.jpg

- Businesslogo.jpg

- The logo for Business Inc.

- A tree with leaves making the logo of Business Inc.

- A tree in the background, its leaves spilling down in the reds, oranges and yellows of the Medieval helmet of the Business Inc. logo

Then, the amount of detail can get to be too much.

- In the background, blurry leaves blend into green grass. A gnarled oak spills its branches out in every direction like a pom pom, a few leaves sticking to the branches still. The gold leaves make up the base of a helmet, red and orange leaves creating the visor and plumage. In a large Serif font, Business Inc is written, a thin double underline cutting through the very bottom of the words.

That description may be good in some contexts, but annoying in others.

- The green grass of a park is blurred, blades indistinguishable from one another. Autumn leaves lay on top, blurred into firey piles at the base of an old tree. Its mostly bared branches stretch in every direction, the few leaves it still has focused on the left side. Their angles seem as if the wind is blowing gently. Next to the tree, with a little space to let the blurry green grass show through, is the helmet of Business Inc.'s logo. Various shaped leaves form together, red spade shaped ones creating the slots of the visor and much of the plumage of the medieval helmet. The other leaves, golden yellow and orange, vary in shapes although most are maple shaped.

That one is just a whole lot of information, and I would only expect it if it was extremely important, like in an art class, to understand the composition.