Impressive OCR

OCR that fits the way you work

Impressive OCR turns scans and PDFs into information you can actually use. Run it as a normal desktop application on your own computer, or install it as a server for automated document processing.

Quick Mode — OCR documents right now

Open Impressive OCR, select one or more files or folders, and start. No project to create, no workflow to configure. It suits scanned letters, invoices and receipts, contracts, reports and publications, existing PDF archives, and images with text in them. Choose the outputs you need and they are all created in one pass.

Pipelines — automate recurring OCR

For document processing that happens regularly, create a pipeline with an input folder and an output folder. Drop a new document into the input folder and Impressive OCR detects it, queues it, processes it, creates the formats you selected and writes them to the output folder.

A document that fails can be retried and then quarantined without stopping the rest of the queue — which is what makes the same software work for a single scanning workstation and for an automated document-processing server.

Server — OCR for your team or an application

Run the same recognition engine headless. A dedicated machine processes documents centrally while people work through the browser, or while other systems feed documents into pipelines. No cloud OCR service is involved at any point.

Two engines, one application

Impressive OCR packages PaddleOCR into an application you install and use, instead of an OCR stack you have to assemble and operate yourself. It ships with two recognition engines, and you pick the one that fits the job.

Accurate is built on the PaddleOCR-VL 0.9-billion-parameter vision-language document model. It is the one for complex documents — magazines, newspapers, multi-column pages, tables, difficult scans, and anything where reading order matters. On a suitable NVIDIA GPU it takes around two seconds a page.

Fast is built on PP-StructureV3 and PP-OCRv6. It is particularly useful for conventional documents such as invoices and forms, and it carries the specialised table, formula, chart and seal recognisers.

How good is the recognition?

Once you know what the product does, accuracy matters. So we tested Impressive OCR against a hand-checked reference transcription of a deliberately difficult German magazine page: two columns, a photograph, complex typography, 637 words.

	Word accuracy	Bag recall	Reading-order loss
Reference transcription	100%	100%	—
Impressive OCR — Accurate	98.4%	98.7%	0.3 pts
Impressive OCR — Fast	95.1%	97.2%	2.0 pts

The Accurate engine differed from the reference in only six places across 637 words. Word accuracy is order-sensitive; bag recall ignores order entirely and measures how much of the page was read at all. The gap between the two is what reading-order damage costs you.

This is one test document rather than a benchmark suite, and we publish it as exactly that: an example of the quality you can expect on a genuinely hard page.

Your documents stay on your machine

Many documents are precisely the files you do not want to upload somewhere else — customer records, contracts, invoices, personnel documents, medical records, internal reports, confidential correspondence.

Impressive OCR processes them locally. Your document content is not sent to a cloud OCR service. There is no cloud OCR account, no API key, no document upload, no per-page cloud processing, and no external copy of your documents made in order to read them. Once the models are installed, recognition runs with no internet connection at all.

For organisations handling sensitive or personal data, that removes an entire external data-processing step from the workflow.

Works with the hardware you already have

A dedicated graphics card is optional. Without one, Impressive OCR runs on the processor alone and the Accurate engine takes roughly 11 seconds a page on the tested system — with the same recognition quality. With an NVIDIA card of 8 GB VRAM or more, the same page takes around two seconds.

	Minimum	Recommended
Graphics card	none — runs on the processor	NVIDIA, 8 GB VRAM or more
Memory	8 GB	16 GB
Operating system	Windows 10 1809+, macOS 12+, Ubuntu 22.04 or equivalent

Free for personal use. Commercial for business use.

The software is not divided into a limited free edition and a more capable paid one. You get the same OCR engines, the same 109 languages and the same output formats either way. What changes is the license.

Personal is free, for private and non-commercial use on up to three machines, under the AGPL-3.0. Both engines, every language, every output format, Quick Mode, watched-folder pipelines, searchable PDFs, local processing.

Commercial is a perpetual license for use in or for a company or other organisation. The same complete software — desktop application, server edition and automated pipelines — with no per-page, per-document or per-seat metering, and perpetual use of the version you bought. It replaces the AGPL’s requirements with commercial licensing terms suited to organisational use.