Review in context
The draft can identify visible content but cannot know the page purpose, surrounding copy, or why the image matters.
Accessibility tool
AI alt text generator
Generate a local first draft, then edit it before copying. The model can describe visible content; it cannot know why the image appears on your page, whether it is decorative, or which name and detail matter to your reader.
or drop one here — JPEG, PNG or WebP, up to 32 MB. It never leaves this device.
Before you download Image captioning
Checking what this device can do with Image captioning…
One-time download
A captioning model reads the image and writes a description of it. It is a ~120 MB one-time download, cached afterwards, and it runs entirely on this device — the image is never sent anywhere.
If you decline: this tool cannot run. Unlike the other AI tools there is no classical shortcut: describing a photograph in words requires a model that has seen photographs.
Choose an image first.
Result proof
A draft that stays editable until you approve it
01
Input
Source image
02
Local pass
Local visual caption
03
Result
Editable text
How to verify it: Read the draft without looking at the image, add the page-specific purpose, remove repetition, and verify any names or text shown inside the pixels.
Text result, not a pixel edit
A draft description stays separate from the canvas
The caption model returns editable text for review. It does not touch the bitmap or create a misleading canvas-history entry, and it does not run until its dedicated model download is approved.
Editable text
Result type
No silent download
There is no classical caption shortcut, so declining consent leaves the image unchanged and explains why no result was generated.
People need careful wording
Check names, identities, emotion, sensitive traits, and uncertain objects before publishing.
Know before you process: Useful alt text communicates the image purpose in its destination context; treat the generated description as a starting point, never an automatic accessibility guarantee.
Good alt text is contextual, not exhaustive
Alt text replaces the image’s job in its surrounding page. A model sees visual objects, but it cannot know whether the same bridge photograph illustrates a location, a lighting technique or a construction detail. The generator solves the blank-page problem; the author still makes the accessibility decision.
What people use this for
Product catalogues
Draft consistent descriptions in batches, then add product names, variants and the detail relevant to the listing.
Editorial photographs
Start from visible facts and add the people, place or event verified by the editor.
Social and CMS cleanup
Fill missing alternatives efficiently while keeping every result editable before it reaches a publishing system.
Accessibility review
Use the draft to spot whether the real requirement is concise alt text, empty alt, or a longer adjacent description.
Limits worth knowing
- The model cannot infer the image’s purpose in the surrounding page.
- Names, brands, locations, charts and text inside screenshots require human verification.
- The current model writes English; translation and local terminology need a separate review.
- A model download is required and there is no pretend rules-based fallback.
How the local workflow runs
- Decide first whether the image is informative, functional or decorative.
- Choose the source image and explicitly approve the local caption model if needed.
- Generate one draft without changing the image.
- Edit for purpose, names, brevity and any text visible in the image.
- Copy the reviewed text, or use empty alt for a decorative image.
Frequently asked questions
Can I publish the generated sentence unchanged?
Treat it as a first draft. Add the purpose, names and context that only the page author knows, then remove details the surrounding text already provides.
Should every image have descriptive alt text?
No. Decorative images should usually use empty alt text so a screen reader can skip them. Functional images need the action or destination, not merely their appearance.
Does the model read text inside images reliably?
No. Dense screenshots, charts and embedded text need human review and often a longer nearby explanation. Verify every word that matters.
Is the image uploaded?
No. The caption model runs locally after a one-time explicit download. The image and edited draft remain in this tab.
Why is there no lightweight fallback?
A useful visual description needs a trained vision-language model. The tool fails clearly when the model is unavailable instead of returning generic or misleading text.
How long should alt text be?
As short as possible while replacing the image’s function in that page. Complex charts often need concise alt text plus a nearby detailed explanation.
Related tools
Ready to try it?
The image stays local and the text stays editable until you approve it.
Draft alt text →