AI: free online tools
5 free tools, processed locally in your browser.
The label on this category deserves qualifying, because the five tools do not work in the same way. Only one of them runs a trained model on your device: extracting text from an image, which uses the Tesseract engine. Translation computes nothing locally and hands the work to a remote service. The other three, the summariser, the rephraser and the title generator, are classical algorithms: frequency statistics, substitution lists and sentence templates.
That precision tells you what to expect. A deterministic process returns the same result for the same text every time, invents no facts and works instantly. In exchange, it will not write in your place.
Tools in the AI category
- Text summariser — Pull out the key sentences of a long text, locally.
- Rephrase text — Shorter, more formal or simpler version.
- Translation — Machine translation via a free API (internet required).
- Title generator — 10 catchy titles for an article or a video.
- Extract text from image — OCR on photos, screenshots or scans.
Summarise or shorten: two tools, two results
The summary is extractive. The text is cut into sentences, stop words are discarded, the frequency of the remaining words is counted, and each sentence is scored on the frequent words it contains. The first sentence gets a bonus, the last a smaller one, and very short sentences a penalty. The best-scoring ones are put back in their original order. What you get, therefore, is sentences from your own text, copied verbatim, never a rewording. The length control aims at roughly 15, 30 or 45 per cent of the sentences.
The rephraser in its Shorter mode does not select: it trims in place. Filler constructions of the "it should be noted that" or "to be honest" kind disappear, intensity adverbs with them, and any sentence whose vocabulary overlaps an already-kept sentence by more than three quarters is dropped. Choose the summary to pull the substance out of a long report, and the shorter rephrasing to slim down a text you want to keep whole.
Rephrasing: what the rules can do
The More formal and Simpler modes apply matching lists in both directions. The first replaces colloquial register with a formal one, French ça becoming cela and boulot becoming travail, then restores the capitals. The second does the reverse, simplifying administrative vocabulary, néanmoins becoming pourtant and afin de becoming pour, and it breaks over-long sentences at their semicolons. Those substitution lists are French, so the two modes only make sense on French copy.
An automatic pass repairs the elisions those swaps break, so that de élément becomes l’élément again. Sentence structure, on the other hand, is left alone: meaning is never rewritten in any depth. To genuinely change the angle of a paragraph, none of these tools will be enough.
Translation and titles: the two special cases
Translation queries the public MyMemory API: your text leaves the browser and passes through that service, which the banner shown above the field states outright. It is cut into pieces of under five hundred characters at sentence boundaries, with a ceiling of 4,500 characters per request. A connection is indispensable, and confidential text has no business here. Around thirty languages are offered, from Mauritian Creole to Japanese.
The title generator calls no service at all: it drops your subject into ten fixed templates, which differ depending on whether you are aiming at an article, a video or a product page, and injects the current year into some of the formulas. The result is a starting point for choosing an angle, not a headline ready to publish — read the grammar and the phrasing before you take one of the suggestions.
OCR, and where to start
Extracting text from an image loads the Tesseract engine and its language model on first use, then recognises the text locally. Six languages are offered, including a French plus English mode for mixed documents, and the result is a text file. Straight, sharp, contrasted framing changes everything; handwriting or a blurred photo will give very little. For a scanned PDF, the dedicated tool in the PDF category is a better fit.
The engine and its language model weigh about fifteen megabytes, downloaded on first use and then cached by the browser, which is why the first run is slow and the following ones are not. A connection is needed once; after that the recognition itself runs on your machine, and your pages are never sent anywhere.
These tools chain together well. A photo of a document becomes text through OCR, is read back through the spell checker, is condensed by the summariser, then translated if the need arises. Note that the summariser wants at least three sentences to work with and never returns fewer than two, so it has nothing to offer on a short paragraph.