AI tools
Extraction and text tools with a little more intelligence.
By opening this website or using its tools, you agree to our Terms and Privacy Policy.
Most popular
All ai tools
Pull text out of an image or scan, in 13 languages.
Copy out the text and tables, with OCR for scanned pages.
Describe a pattern in plain English, or test and explain one you have.
Draft a meta description from your page text, fitted to the SERP width.
Compare two texts, then have the change explained in plain English.
Condense a long PDF into key points.
Diff two PDFs and explain what changed.
About these tools
These are the tools on this site that recognise rather than convert. Everything else here transforms a file whose contents are already known — a HEIC is decoded, a PDF is re-saved. Reading text out of a photograph is a different kind of problem: the letters exist only as arrangements of pixels, and getting them back means a model that has learned what letters look like.
That model runs in your browser. The recogniser is Tesseract, compiled to WebAssembly, with a per-language neural network downloaded once and cached. Nothing is sent to an API, which is unusual for anything labelled AI and is the reason these pages carry the same zero-bytes-uploaded counter as the rest of the site. It also means there is no quota, no account, and no page limit, because there is no server keeping count.
The trade is honesty about accuracy. A cloud OCR service running a much larger model on someone else’s hardware will beat this on difficult input — faint scans, unusual fonts, heavy skew. On the ordinary cases people actually have, which are screenshots, phone photos of documents, and scanned pages, the gap is small and the privacy difference is not. Every result here reports its own confidence so you can see which side of that line your document fell on.
What these tools do not do is understand. They read what is on the page; they do not summarise it, answer questions about it, or check whether it makes sense. Summarising, comparing, and explaining are separate jobs, listed here as planned, and they raise a question these pages do not: those genuinely need a large model, which means either a much heavier download or sending your document somewhere. Nothing will be added to this category that quietly does the latter.
FAQ
Is my file sent to an AI service?
No. Every live tool in this category runs its model inside your browser. The counter under each one monitors the network while your file is processed and stays at zero bytes sent. The engine and language model are downloaded from this site before processing starts, and cached by your browser afterwards.
Why is there a download before the first use?
Because the recogniser has to reach your machine to run there. It is about 3.7 MB of WebAssembly plus 0.6–3 MB for whichever language you choose. That download is the cost of not uploading your document, and it happens once per language rather than once per file.
Which languages can it read?
Thirteen: English, Spanish, French, German, Portuguese, Italian, Dutch, Russian, Simplified Chinese, Japanese, Korean, Arabic, and Hindi. The selection favours script coverage over adding more Latin-alphabet languages, since those share most of their letter shapes anyway. Only the language you pick is downloaded.
Is this as good as Google or Adobe OCR?
On clean input — screenshots, straight scans, well-lit photos of printed text — it is close. On difficult input it is not: a cloud service can run a far larger model than anything sensible to download into a browser tab. The confidence figure shown with every result tells you which case you are in, rather than leaving you to discover it later.
Can it read handwriting?
Not reliably. Tesseract is trained on printed and typeset text. Neat block capitals sometimes work; ordinary handwriting produces confident nonsense rather than an obvious failure, which is worse than no result.
Why are some tools here listed as coming soon?
Because they need a genuinely different kind of model. Recognising letters is a small, downloadable job; summarising a contract or explaining a diff between two documents is not. Those are planned but will only ship in a form consistent with the rest of the site — and if a tool ever does have to send data somewhere, the page will say so plainly rather than in a footnote.