search⌘KGet started

Do barcodes and QR codes come back?

No. txtfetch does not decode them. It reads the printed text near them. Here is what each case returns.

The short answer

No. txtfetch does not decode barcodes or QR codes. The engine reads printed text only. It has no barcode decoder.

The digits printed under a barcode do come back, because OCR reads them as text. They are the caption. They are not the encoded value.

A QR code next to text adds a few stray tokens to the result. An image that holds only a code fails with an error.

The engine is tika-server-standard 3.3.1.

Source: terraform/lambda-layers/build/versions.env (versions.env: TIKA_VERSION), accessed 2026-10-05.

What comes back, case by case

We sent each image to the production API. Every row shows its source. The inputs and the raw responses are in the repo.

What seven barcode and QR code inputs return today
InputWhat comes backSource
A QR code aloneAn error. The image holds no printed text, so OCR finds nothing. The QR code is not decoded, and its payload does not come back.Production API probe: POST /v1/extract
accessed 2026-10-05
An EAN-13 barcode with its digits printed belowThe printed digits come back as text. OCR reads them. The bars are not decoded. Blank lines follow the digits.Production API probe: POST /v1/extract
accessed 2026-10-05
A Code 128 barcode with a captionThe caption comes back, again through OCR.Production API probe: POST /v1/extract
accessed 2026-10-05
The same Code 128 barcode with no captionAn error, the same as for a QR code alone. Nothing is printed, so nothing comes back.Production API probe: POST /v1/extract
accessed 2026-10-05
The same Code 128 barcode with the caption ORDER 9The caption comes back. The encoded value SHIP-4471-AB does not. So nothing reads the bars. Only the printed text returns.Production API probe: POST /v1/extract
accessed 2026-10-05
An invoice image with two text lines and a QR codeThe text lines come back. The QR code turns into a few stray tokens next to them. The payload, PAYLOAD-SECRET-7781, does not come back.Production API probe: POST /v1/extract
accessed 2026-10-05
The same invoice saved as a PDFThe same result. The text lines come back, and the QR code adds stray tokens. Strip them before you chunk the text.Production API probe: POST /v1/extract
accessed 2026-10-05

Real output

Each block is the full response body, as the API sent it. In the JSON, \\n is a line break inside the text.

A QR code alone

In

qr-alone.png

Out

{"status":"error","error":{"code":"extraction_failed","message":"extraction produced no text"}}

An EAN-13 barcode with its digits printed below

In

ean13.png

Out

{"status":"success","extracted_text":"5901234123457\n\n\n","metadata":{"content_type":"image/png","bytes":5513,"chars":16,"ocr":true}}

A Code 128 barcode with a caption

In

code128.png

Out

{"status":"success","extracted_text":"SHIP-4471-AB\n\n\n","metadata":{"content_type":"image/png","bytes":4694,"chars":15,"ocr":true}}

The same Code 128 barcode with no caption

In

code128-nocaption.png

Out

{"status":"error","error":{"code":"extraction_failed","message":"extraction produced no text"}}

The same Code 128 barcode with the caption ORDER 9

In

code128-order9.png

Out

{"status":"success","extracted_text":"ORDER 9\n\n\n","metadata":{"content_type":"image/png","bytes":4154,"chars":10,"ocr":true}}

An invoice image with two text lines and a QR code

In

invoice-qr.png

Out

{"status":"success","extracted_text":"Invoice INV-20931\n\nTotal due: 412.50 EUR\n\nEla]\n[=]\n\n\n","metadata":{"content_type":"image/png","bytes":14858,"chars":53,"ocr":true}}

The same invoice saved as a PDF

In

invoice-qr.pdf

Out

{"status":"success","extracted_text":"\nInvoice INV-20931\n\nTotal due: 412.50 EUR\n\nno)\n\noO\noO\n\n\n\n","metadata":{"content_type":"application/pdf","bytes":25593,"chars":57,"ocr":false}}

What the stray tokens do to your index

The invoice came back with lines such as Ela] and [=] after the total. A chunker keeps them. An embedding model reads them. Run the checks in Quality checks and drop short lines that hold no word.

A failed request is easier to handle. Read the error code in Errors and skip the file or decode the code yourself.

How to read the codes yourself

Decode the code with an open-source library. Run it on the same file you send to txtfetch. The decoder gives you the encoded value. txtfetch gives you the printed text.

  • zxing-cpp. A C++ decoder with a Python package, zxing-cpp. It reads QR codes, EAN, UPC, Code 128, Data Matrix and PDF417. We ran it on all six images. It decoded every code. Licence: Apache-2.0. Read the source → (accessed 2026-10-05)
  • ZXing. The original Java library. Use it when your pipeline runs on the JVM. Licence: Apache-2.0. Read the source → (accessed 2026-10-05)
  • pyzbar, on top of ZBar. A Python wrapper. It needs the ZBar shared library on your machine. The ZBar library has a different licence from the wrapper. Licence: pyzbar: MIT. ZBar: LGPL-2.1. Read the source → (accessed 2026-10-05)

Install

pip install zxing-cpp pillow requests

Python

import os
import requests
import zxingcpp
from PIL import Image

path = "label.png"

# 1. Decode the bars yourself.
codes = [(c.format.name, c.text) for c in zxingcpp.read_barcodes(Image.open(path))]

# 2. Extract the printed text with txtfetch.
with open(path, "rb") as f:
    r = requests.post(
        "https://api.txtfetch.com/v1/extract",
        headers={"Authorization": f"Bearer {os.environ['TXTFETCH_KEY']}"},
        files={"file": f},
    )
text = r.json().get("extracted_text", "")

print(codes)  # [('Code128', 'SHIP-4471-AB')]
print(text)

A decoder reads images. For a PDF, render each page to an image first.

Photos, scans and the premium tier

We did not measure skewed photos, damaged codes or Data Matrix and PDF417 codes. We did not measure barcodes on the premium tier. We make no claim about them.

?quality=premium is sold as a tier, but it does not run today. VLM_ENABLED defaults to false, and no operator has switched the tier on in production. A premium request downgrades to the standard Tika path instead.

Scanned pages go through OCR. Read the image page for the formats it takes.

Frequently asked questions

Does txtfetch decode barcodes and QR codes?
No. The engine has no barcode decoder. It reads printed text only. A QR code or a barcode never returns its encoded value.
Does the number printed under a barcode come back?
Yes, when the image is clear enough for OCR. It comes back as ordinary text. It is the caption, not the encoded value. A label can print a different caption from the one it encodes, so do not trust it as the code.
Why does my barcode image fail with extraction_failed?
The image holds no printed text. OCR finds nothing, so the request fails with the error extraction_failed. A barcode with no caption and a QR code alone both fail this way.
What does a QR code do to the text of an invoice?
The text lines come back. The QR code adds a few stray tokens next to them. They reach your chunks and your embeddings, so check the text before you index it.
How do I read the codes?
Decode them with an open-source library before or next to the txtfetch call. Use zxing-cpp, ZXing or pyzbar. The snippet on this page runs zxing-cpp and txtfetch on the same file.
Does the premium tier read barcodes?
We did not measure barcodes on the premium tier. We make no claim about it. The premium tier does not run today, so a premium request downgrades to the standard Tika path.

Watch it run, then take a key.

The playground replays a real recorded response for every sample document.

Open the playground →

Get an API key →