AI alt text generator

Drop one image or a whole folder and Toolula's on-device AI writes a short alt text for each one plus a longer description — editable, copyable and exportable as CSV or JSON for your CMS or shop.

About this tool

Alt Text and Descriptions Without Sending Your Images Anywhere

Microsoft's Florence-2 vision model runs inside your browser tab, so unpublished product photos and private pictures stay on your device.

Short Alt Text and a Longer Description

Every image gets a one-sentence alt text within the 125 characters screen readers handle well, without filler like “image of”, and a paragraph describing the scene, colours and objects for captions, long descriptions or SEO copy.

Batch Mode With CSV and JSON Export

Add as many images as you like; they are described one after another. Export file name, alt text and description as CSV for spreadsheets and shop imports, or as JSON for scripts and headless CMSs.

Edit, Copy and Translate

Fix any alt text right in the list, copy it with one click, and translate everything into your language with your browser's built-in translator or Toolula's on-device translation models.

How to use it

  1. Drop one or more images, or paste one with Ctrl + V.
  2. Click Write alt text — the AI model loads once (about 170 MB) and then runs on your device.
  3. Check and edit the alt text, copy it, or translate it into your language.
  4. Export all alt texts and descriptions as CSV or JSON.

Frequently asked questions

Are my images uploaded to a server?

No. Every image is decoded, processed and encoded inside your browser tab with Canvas, Web Workers and WebAssembly. Nothing is uploaded, stored or logged.

How long should alt text be?

Short: one sentence that says what matters in the image, usually under 125 characters, because many screen readers read alt text in one go. Leave out “image of” — the screen reader already announces an image. The longer description is for captions or pages that need more detail.

Which AI model writes the text?

Florence-2-base by Microsoft (MIT licence), a compact vision–language model that runs with ONNX Runtime in your browser — on the graphics card with WebGPU, otherwise on the processor. It writes in English; the translate button turns the results into your language.

Why does the first image take longer?

The model is downloaded once (about 170 MB) and cached by your browser. With WebGPU an image then takes a few seconds; without it, on the processor alone, about a minute.

Is the alt text always correct?

No AI is perfect: it can miss text in the image, miscount objects or not know who or what exactly is shown. Read each alt text before you publish it and add names, brands or context the model can't know.

Is it free, and is there a watermark?

Yes, it is free to use without an account, and Toolula never adds a watermark. The free plan has fair-use limits; Toolula Pro, an optional plan, raises them.