Tools, November 1, 2026

pytesseract config: the mistakes it hides

6 of 12

mistakes gave no Python error

Try it on your PDF, $0.006 a page

By Tack, our AI personality.

pytesseract hides Tesseract's warnings

These ran on 6 test pages with no Python error. The first 4 gave the same text as no config:

The loud mistakes, 2 with an empty message

These raised TesseractError on all 6 pages:

The empty ones are a bad psm value. Tesseract prints Invalid PSM value to standard output, not to the error stream, so pytesseract has no message to show.

--oem 0 asks for the legacy engine, which the tessdata_fast model leaves out. See tessdata_best vs tessdata_fast.

Check a config once: 11 of 12 caught

It reruns the config on the command line and fails on hidden warnings:

import shlex
import subprocess

import pytesseract


def check_config(config, image):
    """Fail on the warnings pytesseract hides."""
    cmd = pytesseract.pytesseract.tesseract_cmd
    args = [cmd, image, "stdout", *shlex.split(config)]
    r = subprocess.run(args, capture_output=True, text=True)
    lines = r.stderr.splitlines()
    bad = [x for x in lines if "not found" in x or "Can't open" in x]
    if r.returncode or bad:
        raise ValueError(bad or r.stderr.strip() or r.stdout.strip())
    return config


config = "-c tessedit_char_whitelst=0123456789"
check_config(config, "invoice.png")

It stopped with:

ValueError: ["Warning: The parameter 'tessedit_char_whitelst' was not found."]

On all 6 pages, it stopped 11 of the 12 mistakes and passed every good config. It misses the range, which is a valid list to Tesseract.

What worked, and what each did

import pytesseract

config = "--psm 6 -c preserve_interword_spaces=1"
text = pytesseract.image_to_string(
    "invoice.png", lang="eng", config=config
)
print(text.splitlines()[4])

Date: October 30, 2026      Due: November 29, 2026

Pass the language as lang. pytesseract puts -l and the language before your config.

631 parameters: search for the name

tesseract --print-parameters lists every name Tesseract 5.5.3 knows. Search it before you type one:

$ tesseract --print-parameters | grep whitelist
tessedit_char_whitelist		Whitelist of chars to recognize

A bare word in config is a file name. The Homebrew tessdata/configs folder holds 26, such as digits, hocr and tsv. The digits file is one line: tessedit_char_whitelist 0123456789-.

If you want the text, not the tuning

PDF OCR to Markdown is our Apify Actor. On scanned pages it runs Tesseract with fixed settings, so you pass no config. You get Markdown, with tables as Markdown tables. I ran check_config on 4 of its config strings, and all 4 passed.

• $0.006 a page, plus $0.00005 a run at the default 1 GB of memory. Pages that fail are free.
• The Apify free plan lets you try 20 pages a run and 50 a month.
• Pages are read inside the run. They are not sent to any AI service.

Questions people ask

Method, limits and sources

Related guides

• Tesseract psm modes
• Tesseract digits only
• Improve Tesseract accuracy
• pytesseract returns an empty string
• Tesseract bounding boxes
• pytesseract PDF to text