How it works
The Alt Text Generator audits every image on a page (or on a batch of uploads) and writes a fresh alt-text suggestion per image using an AI vision model.
URL mode: paste any page URL. We fetch the page through our proxy pool, parse the HTML, discover every <img> tag, and for each image: capture surrounding context (page title, H1, nearest preceding paragraph or figcaption within the same <figure>), audit the existing alt for quality issues (missing / empty / generic / duplicate / too-long / keyword-stuffed), and send the image to the vision model with that context.
Upload mode: drag and drop or pick image files (JPEG, PNG, WebP, GIF, up to 10 MiB each). We skip the page-fetch and existing-alt audit; the vision model writes alt text from the image alone. Best for new images you have not published yet.
What the vision model returns per image: a 125-character-or-less alt text suggestion, a decorative-image classification (so we can recommend empty alt for purely visual filler), OCR transcription of any visible text in the image, and a confidence rating (high / medium / low).
Focus keyword (optional): when supplied, the vision model weighs the keyword lightly when it naturally fits the image. Force-fitting keywords that don't match the image is explicitly discouraged in the prompt; we will not stuff keywords on your behalf.
Language: 12 languages supported. Alt text comes back in the chosen language regardless of which language the surrounding page is in.
Plan caps. On the Free plan you can audit up to 5 images per submission and up to 10 images per day. The daily cap exists because each vision call is a paid AI inference billed to our account.
What the per-image card shows
- Thumbnail
- A preview of the image being audited. For URL mode, fetched from the page; for Upload mode, generated client-side via a blob URL so the file never leaves your browser until you submit.
- Existing alt
- The current
altattribute on that<img>tag (URL mode only). Color-coded by quality: green for "good", amber for "too long" / "generic", red for "missing" / "empty on content image" / "keyword-stuffed". - Quality badges
- The detected issues on the existing alt: missing, empty, generic, duplicate, too-long, keyword-stuffed, or good. Each badge is a one-click summary of the problem.
- Generated alt (editable)
- The AI vision model's suggestion. Fully editable in place. The character count + WCAG pass/fail badge updates as you type. Aim for under 125 characters.
- Confidence
- The vision model's self-rated confidence (high / medium / low). Low-confidence outputs deserve more scrutiny before publishing; the model is being honest about cases where the image is hard to interpret.
- Decorative flag
- When the model classifies an image as purely decorative (dividers, brand watermarks), we recommend setting
alt=""rather than describing the image. A "Use empty alt" button is shown. - Text in image (OCR)
- Transcription of any visible text the model found in the image. Useful for infographics, charts with labels, and accessibility audits on text-heavy images.
- Context (URL mode)
- The surrounding text we captured from the page (page title, H1, nearest paragraph or figcaption). What we sent to the vision model as context. Helps explain why a particular suggestion was generated.
Frequently asked questions
What is alt text and why does every image need it?
Alt text (short for "alternative text") is a written description of an image, set on the <img> tag's alt attribute. It serves three audiences: screen readers read it aloud to visually-impaired users, so the alt text IS the image for them. Search engines use it to understand image content for image search and as a contextual signal for the page. Browsers display it when the image fails to load. The W3C Web Content Accessibility Guidelines (WCAG) require alt text on every meaningful image; missing alt is one of the most common accessibility violations on the web.
How do I write good alt text?
- Describe what the image shows, in context. "A golden retriever puppy sitting on grass" is better than "a dog".
- Keep it under 125 characters. Screen readers may cut off longer descriptions; search engines may downweight verbose alts.
- Do not start with "image of" or "picture of". Screen readers already announce "graphic" or "image" before reading the alt; saying it again is redundant.
- Include relevant keywords naturally if they describe the image, but do NOT stuff them. "Running shoes for flat feet on beach" is fine; "best cheap running shoes for flat feet buy now" is keyword stuffing.
- Use empty alt (
alt="") for purely decorative images: dividers, brand watermarks, ornamental flourishes. Screen readers skip them.
Can AI accurately write alt text?
Modern vision models do a strong job describing image content. They reliably identify objects, scenes, basic activities, and visible text. They are less reliable for: brand-specific recognition (knowing this exact logo is YourBrand), nuanced emotions ("delighted" vs "amused"), and context-specific meaning ("this is the same product as the previous photo"). This tool generates a draft alt text per image and surfaces a confidence rating; you should always review and refine before deploying, especially for images where context is critical (product pages, journalism, technical diagrams).
What are decorative images and why is the AI suggesting empty alt for some?
Decorative images carry no information beyond visual flourish: dividers, background patterns, brand watermarks, ornamental icons next to text that already says the same thing. WCAG explicitly requires empty alt (alt="") on these so screen readers skip them, instead of reading "image of decorative divider" to a user who can't see the page. Our vision audit classifies each image and recommends empty alt for the ones that look decorative; you can override before copying.
What do the quality flags on existing alt text mean?
- Missing: the
altattribute is not present at all. WCAG violation. - Empty:
alt="". Fine for decorative images, a violation for content images. - Generic: filename-like or vague ("image", "photo", "DSCF1234.jpg"). The alt exists but tells screen readers nothing useful.
- Duplicate: the same alt text appears on multiple images on the page (signaling lazy copy-paste).
- Too long: over 125 characters. Screen readers may cut off; search engines may downweight.
- Keyword-stuffed: heuristic detection of unnaturally-dense keyword repetition.
- Good: passes all checks.
Why is the focus keyword option there?
If you supply a focus keyword (e.g. running shoes), the vision model is asked to weigh it lightly when it naturally fits the image content. For example, if you upload a photo of running shoes and supply the keyword running shoes, the generated alt will probably read "Running shoes on a wooden floor" rather than just "Athletic shoes on a wooden floor". If the keyword doesn't fit the image at all, the model ignores it rather than force-fitting it (which would be keyword-stuffing). Leave the field blank and the generated alt is purely descriptive.
What's the difference between URL mode and Upload mode?
- URL mode: paste any page URL. We fetch the page through our proxy pool, discover every
<img>tag, capture surrounding context (the nearest<p>, heading, and<figcaption>), audit each image's existing alt, and generate fresh alt text. Best for auditing whole live pages. - Upload mode: drag and drop or pick image files directly. We skip the page fetch and audit (no existing alt to flag); the AI generates alt text from the image alone. Best for new images you have not published yet, or images that live behind login.
Does this work for images with text in them?
Yes. Our vision audit includes OCR (optical character recognition) and transcribes visible text in the image into the per-card output (under "Text in image"). This is particularly useful for infographics, charts with labels, and meme-style images. WCAG also requires that critical information conveyed by text in an image be available either as alt text or as visible page text outside the image, so the OCR output makes that audit fast.