AI Image Describer

The description walks the whole frame in a natural order: the main subject, pose and expression, clothing and props, the setting and background, composition and framing, camera perspective, lighting, colour palette, and mood.

That completeness is the point. It is written for someone who cannot see the picture at all, so nothing important is assumed to be obvious — which also makes it a solid record when you are cataloguing a library.

Everything in it is grounded in what is visibly there. It will not speculate about who someone is, where a photo was taken, or what happened just outside the frame.

Good alt text describes the image's purpose in context, not every pixel in it. A photo of a chart needs the finding, not the axis labels. A decorative background photo often needs nothing at all.

Keep it to a single sentence where you can. Screen readers announce alt text in one uninterrupted run, and a long description becomes hard to hold in your head by the time it ends.

Do not open with image of or picture of. Assistive software already announces that it is an image, so those words are repeated noise on every single item.

So treat the output as raw material rather than a finished answer. Take the opening clause, read it against the surrounding page, and cut anything the context already makes obvious.

People are described strictly by visible appearance — approximate age range, hair, build, expression, pose, and clothing, plus what they appear to be doing.

It never names anyone or guesses who they are, and it will not infer anything the picture does not show. That is what makes it safe to run over photos of real people.

If you are writing a caption from it, the appearance detail is usually the first thing to cut. What the person is doing almost always matters more than what they are wearing.

Descriptions can be returned in any of thirteen languages, and the language is applied when the description is written rather than translated afterwards.

That matters for alt text on a multilingual site. A description composed directly in the target language reads naturally, where a machine translation of English alt text often does not.

Your image is used only to produce the description. It is never stored, never published, and never reused.