ScreenVault

ScreenVault / Screenshot to text

Screenshot to text

Drop or paste a screenshot and get its text back: a stack trace, a chat thread, a table from a dashboard. Edit it, copy it, or save it with the image so your vault can search it.

What it reads well, and where to be careful

Screenshots are the easy case for text recognition: crisp, rendered fonts on flat backgrounds. Expect accurate results on terminal output, IDE code, chat apps, spreadsheets and web pages. Accuracy drops with text under about 9 pixels tall, heavy JPEG compression, text over photos, and very dense tables where columns may be read in the wrong order.

Because a language model does the reading, it understands layout (it will keep a numbered list as a list) but can, rarely, “tidy” something it thinks is a typo. For API keys, invoice numbers and similar values, compare against the image.

Make a folder of screenshots searchable

In the vault, open any screenshot and press “Read text & suggest tags”. The text is stored next to the image in your browser, a title and tags are suggested, and the search box then matches words that only appear inside the picture. That is the practical answer to “search screenshots by text” without uploading your library to a cloud service.

Questions

How does screenshot to text work here?

Your screenshot is downscaled in the browser and sent once to a vision-capable language model (OpenAI GPT-6 Luna via Vercel AI Gateway), which transcribes the visible text. The text streams back into the box. The image is not stored by ScreenVault.

Does it keep code indentation and line breaks?

The model is instructed to keep line breaks, list structure and indentation, and to copy code, URLs and numbers exactly. Always double-check critical values such as IDs or amounts before pasting them anywhere important.

Can it read handwriting or other languages?

It usually reads clear handwriting and most major languages and scripts, and it transcribes in the original language rather than translating. Very small, blurred or low-contrast text may be missed or guessed.

Is there a limit?

To keep this free, each visitor can make a handful of reads every ten minutes, and very large images are downscaled to about 1,800 pixels on the long side first. For long pages, capture in sections.

How do I copy text from a screenshot on my phone without a website?

iOS (Live Text) and Android (Google Lens, “Select text” in Recents) can both select text in images natively. This page is handy on desktop, for code, and for saving the text alongside the screenshot in your vault so search can find it.

Other screenshot tools