Describe what matters, not everything in the frame
Alt text works best when it focuses on what is relevant to the post's point: a product's key feature, the expression on someone's face, the action happening. A flat, exhaustive list of everything visible in the image is worse. A screen-reader user is listening to this description in place of seeing the image, so relevance matters more than completeness.
Skip what the caption already says
If the caption already states the key information, alt text does not need to repeat it word for word. It can focus on visual details the caption does not cover. Redundant alt text wastes a screen-reader user's time without adding information.
Avoid stuffing it with keywords
Alt text written to influence search rather than describe the image defeats its purpose and reads as nonsense to anyone relying on it. Write it for the person who needs it, not as a hidden keyword field.
A short example of the difference
Weak: "photo." Better but generic: "a photo of food." Genuinely useful: "a plated pasta dish with a fried sage garnish, on a dark ceramic plate." The last version gives someone using a screen reader an actual sense of what the image shows.