altloom

Live demo →

An accessibility audit tool that also drafts the fix. Paste in HTML, and altloom finds every image with missing or weak alt text, then writes starting-point alt-text candidates for each one — using an actual trained Markov chain, not a template string.

What it does

1. Audits every <img> tag, entirely client-side, for:

• No alt attribute at all
• Empty alt="" on an image not marked decorative (role="presentation" or aria-hidden="true" — those pass, correctly)
• Alt text that's just the filename repeated back
• Single-word alt text ("chart", "logo", "screenshot") — a general rule rather than a fixed list, so it catches words a hardcoded list would miss
• Alt text too short to carry any real information

2. Drafts alt-text candidates for anything flagged, pulling context from whatever's actually available on the page — a figcaption, a title attribute, or the filename itself (stripped of camera-code noise like IMG_, DSC_, and version numbers) — and runs it through a real order-2 Markov chain trained on alt-text sentence structures. This is the same generative technique behind ribbon_logic and PetalByte, just pointed at a different problem.

3. Lets you pick or write your own. Each flagged image gets three generated drafts as selectable options, plus a fourth: a plain text field to write your own when none of the three fit. This is always available, even for images with zero usable context and zero generated drafts — nothing is ever a dead end.

4. Exports the corrected HTML. Once you've chosen (or typed) an alt-text for each flagged image, one button applies all of it as real alt attributes and outputs the complete corrected markup, ready to copy back into your own code. Images with nothing selected are left untouched rather than guessed at.

5. Light and dark mode, both built around the same flat, solid, labeled-module visual language — not just a color swap. The theme toggle shows the destination (what clicking switches you to), not the current state.

Runs entirely in the browser. No HTML you paste in ever leaves the page, and no external services are called — the whole audit-plus-draft pipeline is local.

What this doesn't do

altloom can't see your images — it only reads text. The drafts it produces are a starting point built from context clues (filenames, captions, existing alt text), not a description of actual image content. Always look at the image and edit before you publish. This is stated plainly in the tool's own UI, not just here.

Stack

Vanilla HTML/CSS/JS. No build step, no dependencies, no framework — open index.html in a browser and it works. Split into index.html, style.css, and script.js.

Running locally

Clone the repo and open index.html directly in a browser. That's the whole setup — there's nothing to install or build.

About

Part of Yafira Martinez's portfolio — https://yafira.xyz