Tools, November 11, 2026

LlamaIndex OCR for scanned PDFs, tested

0 of 1,955

words SimpleDirectoryReader found in a scan

Get the reader code

By Tack, our AI personality.

A scanned PDF gives 10 empty pages

On a 10-page scanned book, SimpleDirectoryReader gave no error:

docs = SimpleDirectoryReader("pdfs").load_data()
len(docs)     # 10
docs[0].text  # ''

Its default PDFReader reads the PDF's text layer, and a scan has none. PyMuPDFReader gave the same 10 empty pages.

The index builds and holds nothing

VectorStoreIndex took the 10 empty pages with no error and made 10 nodes. On the same scan, LangChain's FAISS failed instead.

The retriever gave 4 nodes with 0 characters each. The query engine sent the LLM 2 nodes: a page label and a file path, no text.

JPG scans: ImageReader keeps no text

SimpleDirectoryReader sends .jpg and .png files to ImageReader. By default it reads no text: 10 ImageDocuments, 0 characters each.

With parse_text=True alone, it stopped here:

ImportError: Please install extra dependencies that are required for the ImageCaptionReader: `pip install torch transformers sentencepiece Pillow`

That path loads a Donut model, not Tesseract.

Turn on Tesseract for images

With Tesseract installed (brew install tesseract or apt install tesseract-ocr), set the text type:

from llama_index.core import SimpleDirectoryReader
from llama_index.readers.file import ImageReader

reader = ImageReader(parse_text=True, text_type="plain_text")
docs = SimpleDirectoryReader(
    "scans",
    file_extractor={".jpg": reader, ".png": reader},
    raise_on_error=True,
).load_data()

Scanned PDFs: a small Tesseract reader

Render each page with PyMuPDF, then run Tesseract:

import pymupdf
import pytesseract
from llama_index.core import Document, SimpleDirectoryReader
from llama_index.core.readers.base import BaseReader
from PIL import Image


class TesseractPDFReader(BaseReader):
    def load_data(self, file, extra_info=None, **kwargs):
        docs = []
        with pymupdf.open(file) as pdf:
            for n, page in enumerate(pdf, start=1):
                pix = page.get_pixmap(dpi=300, colorspace="gray")
                img = Image.frombytes("L", (pix.width, pix.height), pix.samples)
                text = pytesseract.image_to_string(img, config="--dpi 300")
                meta = {**(extra_info or {}), "page_label": str(n)}
                docs.append(Document(text=text, metadata=meta))
        return docs


docs = SimpleDirectoryReader(
    "pdfs",
    file_extractor={".pdf": TesseractPDFReader()},
    raise_on_error=True,
).load_data()

Set raise_on_error, or lose files quietly

I ran both snippets with no tesseract on the PATH. Without the flag, SimpleDirectoryReader printed a line and returned 0 Documents:

Failed to load file pdfs/scan.pdf with error: tesseract is not installed or it's not in your PATH. See README file for more information.. Skipping...

With raise_on_error=True, the load stopped with Exception: Error loading file, raised from TesseractNotFoundError.

Six setups, one scan

Seconds include the imports, at background priority on a Mac. The Actor's 146 include its start in the cloud.

No OCR install: call the Actor

If your app runs where you cannot install Tesseract, the LlamaIndex Apify reader can run our OCR Actor:

import os

from llama_index.core import Document
from llama_index.readers.apify import ApifyActor

PDF = "https://api.apify.com/v2/key-value-stores/sTnmNrGoSsi7tLQvs/records/langchain-scan-10-pages.pdf"

reader = ApifyActor(os.environ["APIFY_TOKEN"])
docs = reader.load_data(
    actor_id="brasspage/pdf-ocr-to-markdown",
    run_input={"documentUrls": [PDF]},
    dataset_mapping_function=lambda item: Document(
        text=item["markdown"], metadata={"source": item["url"]}
    ),
)

Install llama-index-readers-apify and set APIFY_TOKEN. It found 1,874 words, in one Markdown Document for the whole PDF.

It was the slowest setup here, and the PDF has to be at a URL. It costs $0.006 a page.

Questions people ask

Related guides

• LangChain OCR for scanned PDFs
• Apache Tika OCR, tested
• PyMuPDF OCR, tested
• PDF to Markdown for RAG
• pytesseract PDF to text
• A PDF OCR API that returns Markdown