Is my file sent to a server?

No. Only the model weights are downloaded. Your images and text are processed inside your browser.

Why is the first run slow?

The model has to download once. After that it loads from the browser cache in a few seconds.

My computer has no WebGPU. Does it still work?

Yes. The model runs on the CPU in a background worker. It is slower but the page stays responsive, and you can cancel at any time.

Are my files uploaded?

No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.

Is it free? Are there limits or watermarks?

Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.

Does it work offline?

Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Home / AI / AI Image Captioner

AI Image Captioner

Write captions and alt text for photos with an on-device AI model, in bulk.

Free, no uploads, no sign-up. Runs 100% in your browser.

How AI Image Captioner works

The model runs with Transformers.js and ONNX Runtime inside a background worker, so the page never freezes. It downloads once from the Hugging Face model hub and is then cached by your browser, so it works offline. With WebGPU it uses your graphics card, without it the CPU. Your files are never uploaded. The captioner is ViT-GPT2, a vision encoder with a small text decoder.

Input: JPG, PNG, WEBP, GIF. Output: CSV, JSON, HTML.

How to use AI Image Captioner

Drop one or more photos.
Pick Fast or Careful.
Press Write captions, then edit any caption you want.
Copy them or download CSV, JSON or ready HTML img tags.

Tips

Careful mode tries several phrasings and picks the best one.
Good alt text describes what matters in context, so add a few words of your own.

AI Image Captioner FAQ

Is my file sent to a server?: No. Only the model weights are downloaded. Your images and text are processed inside your browser.
Why is the first run slow?: The model has to download once. After that it loads from the browser cache in a few seconds.
My computer has no WebGPU. Does it still work?: Yes. The model runs on the CPU in a background worker. It is slower but the page stays responsive, and you can cancel at any time.
Are my files uploaded?: No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.
Is it free? Are there limits or watermarks?: Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.
Does it work offline?: Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Related tools

AI Image Tagger: Tag a batch of photos with your own labels using CLIP zero-shot AI, then export the tags.
AI Object Detector: Find people, cars, animals and 70 more everyday objects in a photo, with boxes and counts.
OCR PDF: Recognise text in scanned PDFs and images in 10+ languages.
AI Summarizer: Summarize PDFs, Word files and long text on your device, as paragraphs or key points.