Loading tool…
Empty alt attributes and file-name alt text are among the most common accessibility failures on the web. This tool drafts a description for every image you add, using a vision-language model that runs on your own device, and lets you edit each one before you copy it or export the lot.
Processing locally on your device
Overview
Alternative text is the written stand-in for an image. Screen readers read it aloud, browsers show it when an image fails to load, and search engines use it to understand what a picture shows. WCAG asks for meaningful alt text on informative images and an empty alt attribute on purely decorative ones. Writing it for dozens of product photos or blog images is tedious, which is why it so often gets skipped.
Captions are written by Florence-2 base, a 230-million-parameter vision-language model from Microsoft released under the MIT licence, running through Transformers.js and ONNX Runtime Web. Each image is scaled to 768 pixels, encoded by the vision tower and decoded into text. Florence-2 has three captioning modes, used here as the three lengths: a one-sentence caption for alt text, a detailed caption of two or three sentences, and a full paragraph that mentions colors, secondary objects and the background. On a graphics card with WebGPU the model uses half-precision vision weights (about 360 MB); on the CPU it uses 8-bit and 4-bit weights (about 230 MB) and takes roughly 5–8 seconds per image on a typical laptop.
If Florence-2 doesn't fit in your device's memory, the tool switches to DistilViT, a 157 MB Apache-2.0 captioner from Mozilla that writes one short sentence in about a second. You can also choose it directly for quick drafts. Small on-device models describe what they see but don't know why the image is on your page, occasionally misidentify objects, and are not built to recognize people. That is why every result is editable, an optional prefix can add the product name or context, and a filler filter removes openings like “In this image we can see” before you review the text.
| Length | Model task | Typical output | Best for |
|---|---|---|---|
| Alt text | <CAPTION> | One sentence, ≤125 characters | alt attributes, product images |
| Detailed | <DETAILED_CAPTION> | Two or three sentences | Captions, social media descriptions |
| Full | <MORE_DETAILED_CAPTION> | A short paragraph | Long descriptions, complex photos |
| Fast model | DistilViT caption | One short sentence | Quick drafts, low-memory devices |
Drop one image or a whole batch (JPG, PNG, WebP, GIF).
Pick a length: Alt text, Detailed or Full description.
Optionally add a prefix such as the product or page name.
Click Describe and wait for the model to download once and caption each image.
Review and edit each description; the counter warns when alt text passes 125 characters.
Copy single results or <img> tags, or export everything as CSV, JSON or HTML.
Use cases
Draft alt text for a catalog of product photos in one pass, add the product name with the prefix field, then fix colors and variants the model can't know. Export CSV to paste into your store's bulk editor.
Before publishing, describe every image in an article and copy the finished <img> tags straight into your CMS's HTML view.
When an audit finds hundreds of images with empty or meaningless alt attributes, generate first drafts for all of them so reviewers only have to edit, not write from scratch.
Use the detailed length to get a fuller image description for platforms that support alt text on posts, then trim it to what matters.
Pro tips
Answers to common questions about the caption / alt text.
It's better not to. The model describes what is visible but doesn't know the purpose of the image on your page, and it can misidentify objects. A quick review turns a generic caption into useful alt text.
Alt text uses Florence-2's caption mode for one sentence, trimmed to 125 characters if the limit is on. Detailed asks for a two- or three-sentence description, and Full asks for a paragraph that covers colors, secondary objects and background.
Florence-2 needs roughly 1.5 GB of memory on the CPU path. If your browser can't provide that, the tool falls back to DistilViT, which writes shorter, simpler captions but runs almost anywhere.
No. The models describe visible scenes, objects and actions and do not recognize who someone is. Add names yourself through the prefix or while editing.
Search engines use alt text to understand images, which can help them appear in image search. Write it for people first; stuffing keywords into alt text hurts accessibility and brings no real benefit.
No. ToolEnhancer processes files on your own device inside the browser. They are never sent to, stored on or seen by our servers.
Yes. The caption / alt text is free to use with no sign-up, no watermarks and no daily limits.