Alt Draft

Alt text, drafted for you.
Finished by you.

Drop an image. An AI in your browser writes a first draft. You edit it and make the call.

Drop an image here

You can also paste an image from your clipboard.

Your image never leaves your device.

How it works

  1. You drop in an image.

    It opens on your device. Nothing is uploaded.

  2. An AI in your browser writes a first draft.

    The model is LFM2.5-VL by Liquid AI, run with Transformers.js on your own graphics chip. The first visit downloads it (about 550 MB); your browser keeps a copy for next time.

  3. You finish it.

    The draft is a starting point, not finished alt text. It's a small model: it can misname things, misread text or miss the point of the image. It tries to name famous fictional characters and can get them wrong; it does not name real people from their faces. You know why the image is there, so you make the call.

Good alt text is short, says what matters for the page it's on, and quotes any important text in the image. If an image is only decoration, mark it decorative and leave the alt empty.

Alt Draft needs a browser with WebGPU: a recent Chrome or Edge, Safari 26 or later, or Firefox 141 or later on Windows.

Privacy

Your image never leaves your device. The AI runs inside this browser tab, so there is no server to send it to.

The code is open: github.com/TarunDaBoss/alt-draft