AI alt text generator

What alt text is actually for

Alt text is the sentence a screen reader speaks aloud when it reaches your image, and the sentence a search engine reads when it tries to work out what your page is about. It exists for people who cannot see the image — because of a disability, a slow connection, or a broken file path.

That is a narrow, specific job, and most alt text fails at it in one of two ways. Either it is missing entirely, or it is stuffed with keywords for a search engine that stopped rewarding that a decade ago. Both leave the same person staring at nothing.

The reason alt text goes missing is not that people disagree with it. It is that writing three hundred descriptions is a joyless afternoon, and the payoff is invisible to the person doing the work. So it gets postponed, then shipped without, then forgotten.

What this tool does about it

PicsTag looks at each image and writes a plain description of what is in it — the first draft you were never going to write by hand. You then do the part a model genuinely cannot do: decide whether that description says the right thing for this page.

Three ways to feed it:

• A folder of images. Drag them in. Works for a hundred at a time.
• A CSV of URLs. If the images are already live on your site, export their URLs into a CSV with an image_url column and an asset_id column, and PicsTag will fetch and describe each one.
• One image. Sometimes you just need a sentence for the hero image and you do not want to write it.

Then export to CSV or JSON, and paste the descriptions into your CMS — or hand the file to whoever owns the CMS, which is usually the actual bottleneck.

Alt text the model gets right, and where it doesn't

An image captioning model is good at the literal layer: it will tell you there is a woman in a red coat crossing a street in the rain. That is often exactly what you need, and it takes it three seconds instead of your thirty.

What it cannot know is why the image is on the page. The same photo needs different alt text in a news article about a storm ("pedestrians crossing a flooded street in central Manchester") and in a product page for a raincoat ("model wearing the Aurora parka in red"). Alt text describes the image'sfunction, not just its contents, and the function lives in your head, not in the pixels.

So use the output as a draft, and spend your editing time on the images that carry meaning:

• Charts and diagrams — the model will describe a chart, not what the chart says. Rewrite those.
• Images with text in them — the text is the content. It belongs in the alt.
• Product images — the model does not know your product name, colourway or SKU. It knows it is a shoe.
• Decorative images — these need an empty alt="", not a description. Delete the draft entirely.

A rule of thumb for good alt text

Read the page out loud, and when you reach the image, say the alt text instead. If the sentence you just said was useful, it is good alt text. If it was a list of keywords, or a description of something that did not matter, it is not.

Beyond that: keep it under about 125 characters (screen readers do not chunk longer strings well), do not start with "image of" (the screen reader already said that), and do not repeat the caption that is already sitting right underneath the image.

More on this in how to write good alt text, including the cases where the right answer is to write nothing at all.

Why it runs in your browser

Every other AI alt text tool works by uploading your images to a server. That is fine for a blog post about holidays. It is a real problem for an agency handling client assets under NDA, an e-commerce team with unreleased product photography, or anyone whose data processing agreement does not list a new AI vendor.

PicsTag has no server to upload to. The models are downloaded once (about 250 MB, then cached by your browser) and run in a Web Worker inside your tab. You can verify it: open your browser's developer tools, watch the Network tab while you process a batch, and you will see nothing leave.