Perfect product alt text: the AI that looks at the photo and knows its name
AISA now describes images right from the Media Library: it actually looks at the picture, weaves the product name into the sentence, respects the media language — and if you don't like the result, one click takes you back.
The examples in this article are not mockups: they are descriptions actually generated while testing on a real e-commerce catalog — thousands of images in five languages. The feature ships with AISA v1.99.915.
The average catalog is full of "IMG_1234.jpg"
Alt text is what describes an image to whoever can't see it: screen readers, Google Images and — more and more — the AI engines reading your pages. On an e-commerce site it counts double: it's an SEO signal and an accessibility requirement. Yet in most catalogs the alts are empty, or they're just the file name: coat-rack-5.jpg, DSC_0042, screenshot-2024.
The reason is simple: hand-writing a thousand alt texts is a job nobody will ever do. And tools that generate them from the file name alone produce polite garbage.
AI that looks at the photo — and knows your catalog
AISA Vision does two things at once: it actually looks at the image with a vision model (Claude, GPT or Gemini — your own keys or Cloud credits) and it cross-references what it sees with the context. If the photo belongs to a WooCommerce product, the product name enters the sentence naturally — not glued in front like a label.
Before Just the file
"coat-rack-5.jpg" — empty alt, or "coat rack 5" at best. To Google and to a screen reader, that image doesn't exist.
With AISA Vision The scene, with the right name
"Alfa & Beta, five-arm coat rack on a grey wall in a modern room, with a vase of dried flowers." The product is there, the scene is there, the sentence reads well.
And names stay names: "Maestrale" never becomes "Mistral" in the English alt. AISA Vision respects the Brand Context "do not translate" glossary (Translation Fidelity) — the exact spot where machine translation ruins catalogs. Machine names, on the other hand, are ignored: if the file is called IMG_1234, the AI simply describes what it sees.
Single image, selection, or the whole site
From the alt field
The "Describe with AISA Vision" button sits under the Alternative Text field, in the Edit Media screen and in the details modal. One click and you choose: alt only, title only, or both.
From the Media Library
Grid → Select → tick the images → Describe. It processes them in sequence with live progress on the button, and tells you how many it described at the end.
Bulk, One-Click and Scheduler
The same engine powers the image Bulk, One-Click and scheduled regeneration: thousands of alts in the background, with the same contextual intelligence.
A Spanish media file gets described in Spanish
With WPML or Polylang, the declared language of the media decides the language of the description: the English duplicate of a photo gets an English alt, the German one gets German — even when the image isn't attached to any post and the title is a machine name. No more multilingual catalogs with every alt in the wrong language.
The Undo nobody else has
Every run saves the previous values. If the result doesn't convince you, "Restore previous alt" appears next to the button for a single image, and "Undo last description" in the Library puts the whole batch back — alts and titles, only the fields that were actually touched. No other SEO plugin has undo on AI descriptions: you can experiment without fear of ruining a curated catalog.
- Choose what to rewrite — alt only, title only, or both: a hand-crafted title is never touched without your OK.
- See the result — the description lands in the field, already saved.
- Don't like it? — one click on Restore and everything is back. Try again, maybe with a different vision provider.
Try AISA Vision on your catalog
Alt text that truly describes, with the right name, in the right language — and reversible. Included in AI SEO & GEO Assistant, with your own API keys or Cloud credits.
AI SEO & GEO Assistant · The quoted descriptions are real AISA Vision output (v1.99.915) generated during testing; product names shown are fictional. Quality also depends on the chosen vision model.