Image to Text: Get the Words Out and Get On With It
A build fails. The error is forty lines of stack trace in a terminal on a machine you can't copy from — a remote console, a locked-down VM, a colleague's screenshot pasted into chat. You can read it perfectly well. You just can't select it, and retyping a Java package path by hand is how typos get invented.
That is what this page is for. Upload the screenshot and the text in it is recognised and returned as plain text you can paste into a search, an assistant, or an issue. Real OCR on the pixels, not a description of the file. It reads JPG, JPEG, PNG, GIF, BMP, TIFF, TIF and WebP. Free, no sign-up, 50 MB per file, nothing stored afterwards. It's the image to text converter online free case in its most literal form: put a picture in, get the words out.
The Screenshot-to-Search-Box Loop
The most common reason people need to extract text from a screenshot is that they want to put that text somewhere that only accepts text. Search engines don't take images. Neither does your issue tracker's search, your log aggregator, or the box where you'd paste an exception to ask a model what it means.
So the loop is: screenshot, convert, paste. Ten seconds, and the difference is real — an assistant given the exact exception class and line numbers gives a specific answer, while one given a vague paraphrase gives a generic checklist. The same applies to error dialogs, CI output, database messages, and crash reports. If it appeared on a screen as characters and you couldn't select it, this is the way round that.
Plain Text or Markdown? The Honest Answer
Most sites would invent a dramatic distinction here. The truth is less dramatic. On a simple image — a screenshot of a paragraph, a terminal window, a note — plain text and Markdown come back effectively identical. Recognised text arrives as lines, and there is no hidden structure in a block of prose for Markdown to add. Choose either one and you lose nothing.
Where it starts to count is when the picture shows something with a shape: a table, a form, a document page with headings. Markdown can carry that arrangement; flat text can only hint at it with spacing. For those, image to Markdown is the better page. For everything else — and "everything else" is most screenshots — plain text is the cleaner artefact, because there are no syntax characters to strip before it goes into a script, an index, or a diff.
The format toggle at the top swaps between the two and keeps the file you already selected, so trying both costs one upload rather than two.
Everyday Jobs This Quietly Solves
- Terminal and log output. Command output from a session you can't scroll back through, or a log window captured before it rotated away. Get it into text and you can grep it, diff it, or ask about it.
- Chat and email screenshots. Conversations forwarded as pictures instead of text, which is most of them. Convert once and you have something quotable and searchable.
- Signage, labels and packaging. A photo of a serial number plate, a wiring label, a product spec panel. Awkward to retype, trivial to convert.
- Handwritten notes. Worth a try, with realistic expectations — neat printing reads much better than fast cursive, and you should always check the result against the original.
- Code you can only see, not copy. A snippet in a video frame, a slide, or a locked document. Pull it out as text, then pair it with your GitHub repository as text or a GitLab project so an assistant can see the snippet and your codebase together.
- Anything you screenshotted off a website. Habit worth breaking — Web2Txt fetches the page directly and keeps the links intact, and HTML to Markdown handles a saved page. Going back to the source always beats reading pixels.
Where the Recognition Runs, and Where It Doesn't
There's one rule that explains every failure mode worth knowing about: OCR works on the image file you upload, not on images buried inside other files. A picture riding inside a bigger container gets skipped, because the converter is reading the container.
- A scanned PDF returns nothing. An image-only PDF has no text in it, and OCR doesn't reach through the PDF wrapper to the pixels. Fix it at the source: open it in Acrobat, Preview on macOS, or almost any modern PDF reader and run "recognise text" to make it a searchable PDF, then use PDF to text or PDF to Markdown. That's the proper home for scanned documents.
- Images inside a ZIP are listed, not read. Archive conversion gives you the filename and nothing more. ZIP to text is the right tool for a folder of documents; for pictures, extract them and upload each image by itself.
- Images embedded in Word, PowerPoint, RTF, EPUB or email aren't read. They come through as an image placeholder in the output. A DOCX assembled from screenshots converts via Word to Markdown into the surrounding text with holes where the pictures were. Save the images out and convert them here.
Carry that model around and you'll never waste an upload: to read a picture, hand over the picture.
Doing It at Volume, and Feeding Machines
Plain text is the format you want when the output isn't destined for a human. Recognised text from a stack of images concatenates cleanly, chunks predictably, and embeds without carrying formatting noise into the vector — which is exactly what you need when you're building a search index over a pile of screenshots, or turning photographed archive material into something retrievable.
The practical route is one image per upload, since archives skip images, so plan for a loop rather than a single batch. The token counter under the output is useful here: it tells you how much each conversion actually costs before you start stacking pages into one prompt or budgeting a chunking strategy. For the mixed corpora — documents alongside the pictures — File2Txt handles every supported format from one place, and a local directory conversion flattens the text-based side of a project in one pass.
Give the Recogniser an Easy Time
Accuracy follows image quality closely enough that a few habits change your results more than any setting would:
- Take a screenshot rather than a photo of a screen. No glare, no angle, no blur, no moiré pattern from the display.
- Crop tight to the text you want. More pixels per character is the whole game, and it also stops the output filling up with interface chrome you didn't ask for.
- Keep the image the right way up and roughly square-on. Skewed lines are harder to read than straight ones.
- Avoid the hard cases where you can: low-resolution captures, text over photographs, script and decorative fonts, and anything half in shadow. These can work, but they're where errors cluster.
- Proofread anything that matters. A misread digit in a reference number is quiet and expensive, and a five-second glance at the output catches it.
Frequently Asked Questions
How do I convert a JPG to text?
Drop the image into the converter above. OCR runs on the actual pixels and the recognised words come back as plain text you can select, search and paste. Free, no sign-up, 50 MB per image. This is the literal jpeg to text converter case — picture in, words out, nothing to install.
Does it work with PNG and other image formats?
Yes: JPG, JPEG, PNG, GIF, BMP, TIFF, TIF and WebP all work. PNG is the format most screenshots arrive in and it is the best case for OCR, because a screenshot is rendered text rather than a photograph of paper — no camera blur, no page curl, no uneven lighting to work around.
Can it read handwritten notes?
Badly, and you should assume no. OCR models are trained overwhelmingly on printed type; cursive and hurried block capitals produce output that looks plausible and is wrong in the specific places you would not think to check. Neat printed capitals on ruled paper sometimes come through. Anything else needs proofreading against the original, which usually costs more than retyping.
Why did my image come back with the wrong characters?
Resolution, nearly always. OCR needs roughly 300 DPI equivalent to separate similar shapes, and below that rn collapses into m, 1 and l swap, and 0 becomes O. A phone photo taken closer, or a screenshot captured at full window size rather than scaled down, fixes most of it.
Can I get text out of a photo of a document page?
Yes, and it works better than people expect — but flatten the page first. Hold the camera parallel to the paper rather than at an angle, get even light with no shadow across the text, and fill the frame with the page. Keystoned or shadowed photos lose whole lines, and OCR reports no error when it does.
Is the image stored anywhere after conversion?
No. It is sent for recognition, processed, and discarded once the text is returned — not archived, not used for training, not indexed. There is no account attached to it and no copy retained on our side.
Everything Else Here
Images sit alongside twelve other supported formats, all reachable from File2Txt. There's a longer write-up on getting files ready for LLMs if you want the general argument rather than the image-specific one.
Repo2Txt is built and maintained by v12hero, an independent developer building privacy-first native and web apps.