Extract text from an entire folder
Drop images or a folder — text extracted from each via Tesseract.js (21 languages). Download as combined .txt or .zip of individual files. No server.
Drop a folder of photos
or pick a folder / files / ZIP
or paste (Ctrl+V)
Up to 500 files — JPEG, PNG, WebP, GIF, AVIF, HEIC, BMP, SVG, ICO, TIFF — or a ZIP archive
Why batch extract text from images
When you have dozens or hundreds of scanned pages, receipts, or screenshots, copying text one by one is slow. Batch OCR lets you point at a whole folder and pull text from up to 500 images in one pass, so you can digitize archives, feed search indexes, or prep documents for translation without manual retyping.
- 1Select up to 500 images (JPEG, PNG, WebP, GIF, AVIF, HEIC, BMP, SVG, or TIFF).
- 2Pick the recognition language from 21 supported options.
- 3Choose an output mode: combined .txt or individual .zip.
- 4Start the run; the tool processes files sequentially with one reused worker.
- 5Watch per-image confidence scores from 0 to 100 as they finish.
- 6Download the combined .txt or the .zip containing one .txt per image.
- Supports up to 500 images per batch with any mix of supported formats.
- Runs Tesseract.js v7, the same engine as the single-image OCR tool.
- 21 languages available, selectable per batch.
- Two output modes: a combined .txt with --- filename.ext --- headers, or a .zip of individual .txt files via JSZip.
- Each image reports a confidence score from 0 to 100; failed items are marked as error.
- •Digitize folders of scanned documents into searchable text.
- •Archive stacks of receipts and invoices in one pass.
- •Convert batches of screenshots into editable notes.
- •Prepare sign and menu photos for translation at scale.
- •Turn research notebooks and clippings into text you can search.
- •Make batches of image-only PDFs accessible to screen readers.
Recognition quality rises with input quality. Keep images sharp, well-lit, and oriented correctly, and the same engine that handles single OCR will handle your batch reliably.
- •Use high-resolution inputs; aim for at least 150-300 px per character height.
- •Ensure strong contrast between text and background.
- •For very large batches, split into chunks of a few hundred files.
- •Pick the right language for the text in the images.
- •Crop images to the text area to speed up recognition.
All recognition happens in your browser using WebAssembly, so your images never leave your device. The only network request is for the Tesseract language model, which is fetched from a CDN and cached for reuse.
- •OCR runs fully in-browser via WASM.
- •No image is uploaded to any server.
- •The language model is loaded from a CDN and cached locally.
- •ZIP packaging is done locally with JSZip, entirely client-side.
Frequently Asked Questions
Find answers to the most popular questions about our service, tools, and features. If you didn't find what you need — write to us.
It extracts text from a folder of up to 500 images at once using OCR. Recognition runs in your browser via Tesseract.js, and results are delivered as a single combined .txt or as individual .txt files inside a .zip.
Tesseract.js v7 is the same OCR engine used by the single-image tool. It loads a WASM core and a trained language model, then recognizes text in each image pixel-by-pixel and outputs recognized characters with confidence scores.
21 languages, including English, Russian, Spanish, Chinese, Japanese, Korean, Arabic, Hindi, French, German and more. Each language requires its own model that is downloaded from the CDN on first use and cached.
Two modes. Combined .txt merges every image into one text file separated by `--- filename.ext ---` headers. Individual .zip packs one .txt per image (via JSZip) so each result stays separate.
Yes — MAX_FILES is 500. You can drop a whole folder of up to 500 images and the tool will process them sequentially.
To keep memory low (~50-80MB) and the browser responsive, the tool reuses a single Tesseract worker and processes images one at a time. Parallel OCR on hundreds of images would spike RAM and CPU; sequential keeps it stable on any device, including mobile.
For each image the engine reports a confidence value from 0 to 100, estimating how sure it is about the recognized text. Low confidence usually means low image quality, unclear fonts, or unsupported characters.
JPEG, PNG, WebP, GIF, AVIF, HEIC, BMP, SVG and TIFF. Most browsers decode the common ones natively; HEIC and some others may need a browser that supports them.
Related tools
Continue with tools that pair well with this one
