pdftopdfa

pdftopdfa is a free and open-source alternative to Ghostscript-based PDF/A converters. Ghostscript uses a dual license (AGPL/commercial) that makes it difficult to use in commercial products without purchasing a license. pdftopdfa uses MPL-2.0-or-later, a file-level copyleft license. It permits commercial use and combination with proprietary code, provided its terms are followed. For non-OCR conversions, pdftopdfa modifies the PDF structure directly using pikepdf (based on QPDF), avoiding full-document re-rendering and preserving the original content, fonts, and layout where possible.

Highlights

• No Ghostscript required -- direct PDF manipulation via pikepdf/QPDF
• PDF/A-2a, 2b, 2u, 3a, 3b, 3u -- supports modern PDF/A levels (ISO 19005-2 and ISO 19005-3), including Tagged PDF output for scanned documents
• PDF/UA-1 -- optional dual-conformance output with PDF/A-2a or PDF/A-3a
• Auditable PDF/UA workflow -- validated candidates, original-input fallback, explicit machine and author-review states, and atomic JSON evidence reports
• Automatic font embedding -- uses policy-approved Windows system fonts or bundled replacements
• Font subsetting -- reduces file size by removing unused glyphs
• CJK support -- embeds Noto Sans CJK for Chinese, Japanese, and Korean text
• ICC color profiles -- automatically embeds sRGB, CMYK, and grayscale profiles
• XRechnung metadata -- adds canonical Factur-X XMP metadata for recognized, unambiguous embedded XRechnung 3.0 invoices in PDF/A-3 output
• Batch processing -- converts entire directories, optionally recursive
• Integrated validation -- checks conformance via veraPDF
• OCR support -- optional PP-OCRv6 Medium text recognition on the CPU or through DirectML in the supported Windows 11 configuration, with external offline text-model directories, a bundled page-orientation model, and no runtime model downloads
• Layout-aware OCR -- optional column-based reading order for multi-column documents without an additional model or OCR pass
• Table recognition -- recognizes already-cropped bordered ("wired") and borderless ("wireless") tables as typed cells and HTML using only explicitly supplied local ONNX models
• Simple API -- usable as CLI tool or Python library

How It Works

pdftopdfa applies a multi-step conversion pipeline to make a PDF compliant with the PDF/A standard:

1. Pre-check -- copies encrypted and, by default, digitally signed PDFs unchanged, and otherwise detects if the PDF is already a valid PDF/A file (skips conversion if the existing conformance level meets or exceeds the target within the same PDF/A part; optionally skips any veraPDF-compliant PDF/A via --skip-any-pdfa; see the Usage Guide for details)
2. OCR (optional) -- optionally orients pages with the bundled PP-LCNet document-orientation model, straightens only scan-like raster-dominant pages, and recognizes text with externally supplied PP-OCRv6 Medium models; OCRmyPDF rasterizes OCR target pages and creates the searchable text layer
3. Font compliance -- analyzes all fonts, embeds missing ones, adds ToUnicode mappings, subsets embedded fonts, and fixes encoding issues
4. Sanitization -- removes or fixes non-compliant elements (JavaScript, non-standard actions, transparency groups, annotations, optional content, etc.)
5. Metadata -- synchronizes XMP metadata with the document info dictionary, sets the PDF/A conformance level, and optionally adds PDF/UA-1 identification with the required PDF/A extension schema
6. Color profiles -- detects color spaces and embeds the required ICC profiles (sRGB, CMYK/FOGRA39, sGray)
7. Logical structure -- for level A, preserves and safely repairs valid rich tags or builds headings, paragraphs, lists, tables, figures, artifacts, links, forms, and reading order from final digital-PDF provenance or the internal OCR engine's line and layout data. A provably inverted single-column block order is rebuilt; ambiguous or multi-column existing orders are preserved
8. Save and validate -- stages the output with the correct PDF version header, validates requested profiles, and by default atomically publishes only an accepted candidate

Installation

Prerequisites

• Python 3.12, 3.13, or 3.14
• macOS 14 or later on Apple Silicon, Linux, or Windows

Intel-based Macs are not supported. CPU OCR on macOS requires the ARM64 wheels provided by ONNX Runtime for Apple Silicon.

python -m pip install pdftopdfa

If the pdftopdfa console script is not on PATH, use python -m pdftopdfa in the examples below.

PDF/A and PDF/UA validation

Validation uses the external veraPDF application, which is not bundled. Install version 1.30.2 or newer and make its launcher available on PATH, or set VERAPDF_PATH to the executable or its parent directory. --validate and validate=True opt ordinary PDF/A output into validation. PDF/UA output always attempts validation against both the selected PDF/A profile and veraPDF's ua1 profile. If a requested check cannot run or fails, the staged candidate is withheld and the original input is copied to the output instead. Processing errors in OCR, fonts, sanitization, tagging, or saving use the same fallback. The result reports skipped=True, level=None, and target_produced=False with a warning; the unchanged copy is not a converted PDF/A. If even copying is impossible, a failed result is returned and batch processing continues. Use --publish-noncompliant or publication_policy="always" only when an explicitly non-conforming review candidate is required.

Passing --audit-report report.json writes an atomic, machine-readable record containing the exact staged-candidate hash when available, sanitizer and tagger counters, structured veraPDF rule findings, and raw XML for completed validator runs. A fatal invocation replaces stale evidence with a structured fatal_error record. The report distinguishes machine validation from author review; it is evidence, not an accessibility certification.

Optional: OCR support

Install exactly one OCR runtime. CPU inference is the default:

python -m pip install "pdftopdfa[ocr]"

For the supported DirectML configuration on Windows 11:

python -m pip install "pdftopdfa[directml]"

Do not install both extras in the same Python installation: onnxruntime and onnxruntime-directml provide overlapping runtime files. pdftopdfa supports DirectML on Windows 11 with a DirectX 12-capable integrated or dedicated Intel, AMD, or NVIDIA GPU and a current graphics driver.

OCR uses PaddleOCR 3.7 with the selected ONNX Runtime provider. Installing the DirectML extra does not select it automatically; use --ocr-execution-provider directml or ocr_execution_provider="directml". CPU remains the default. If DirectML is requested but unavailable, the OCR runtime reports an error without switching to the CPU. The high-level converter preserves the original input and reports the cause as a warning.

On a machine with several GPUs, directml:<index> passes a raw DXGI adapter index, for example --ocr-execution-provider directml:1. Plain directml uses DirectML's default adapter. The internal diagnostic helper pdftopdfa._ocr_runtime.list_directml_devices() lists the available adapters and their raw indices; as part of a private module, it has no public API stability guarantee. The indices may have gaps because software adapters are omitted, and repeated DXGI entries with the same PCI identity are listed once using their lowest index. Use the reported index, not its position in the filtered list.

The page-orientation model is bundled. PP-OCRv6 text-recognition and table models are external and are never downloaded at runtime. Pass their local directories to each top-level conversion or recognition call; an OCRSession instead receives the PP-OCRv6 text-model pair once when it is created and reuses it across its image-recognition calls. CPU and DirectML use the same FP32 ONNX model files. See the OCR guide for the recognize_table() model contract and typed result. Cell text and grid structure come from the table-structure model, while bounding boxes come from the separate cell-detection model; if the two models report different cell counts, cells are returned without bounding_box and confidence instead of failing.

PP-OCRv6 model setup

The following model revisions are tested and recommended:

• Detection: PP-OCRv6_medium_det_onnx at 6132380
• Recognition: PP-OCRv6_medium_rec_onnx at 50c7eac

Each model directory must contain exactly inference.onnx and inference.yml. Before initialization, pdftopdfa performs a quick structural check that rejects missing or extra entries, non-regular files, and symbolic links. PaddleOCR then loads the model files and checks that they are compatible detection and recognition models. pdftopdfa does not verify the repository revision or model-file hashes.

The models are not included in the source distribution or wheel. Keep them in deployment-managed, read-only directories. Both --ocr-detection-model-dir and --ocr-recognition-model-dir are required together; supplying the pair enables OCR without an additional --ocr flag. Conversely, --ocr, --ocr-force, --deskew, --rotate-pages, --ocr-layout, --ocr-figure-text, and a non-CPU --ocr-execution-provider value (directml or directml:INDEX) are rejected unless both model options are present.

--ocr-lang defaults to en. Use de for German and de+en for mixed German/English recognition. Latin-script languages restrict decoding to Latin letters while retaining numbers, punctuation, and symbols, which prevents Chinese-character output on German scans. The accepted PaddleOCR 3.7 codes are:

af, az, bs, ca, ch, chinese_cht, cs, cy, da, de, en, es, et, eu, fi, fr, french, ga, german, gl, hr, hu, id, is, it, japan, ku, la, lb, lt, lv, mi, ms, mt, nl, no, oc, pl, pt, qu, rm, ro, rs_latin, sk, sl, sq, sv, sw, tl, tr, uz, vi.

Legacy codes such as eng and deu are not accepted. See the PaddleOCR language documentation for the language families represented by these codes.

Quick Start

# Simple conversion (creates document_pdfa.pdf)
pdftopdfa document.pdf

# Specific PDF/A level
pdftopdfa -l 2b document.pdf

# Tagged PDF/A-2a output with veraPDF validation
pdftopdfa -l 2a --validate document.pdf

# PDF/A-2a and PDF/UA-1 output (both profiles are always validated)
pdftopdfa -l 2a --pdfua --document-title "Annual report" \
  --document-language en-GB --audit-report document-audit.json document.pdf

# Explicitly retain a failed validation candidate (not conforming output)
pdftopdfa -l 2a --pdfua --publish-noncompliant document.pdf

# With validation (note: -v = --validate, not verbose; use --verbose for logs)
pdftopdfa -v document.pdf

# Skip any existing veraPDF-compliant PDF/A
pdftopdfa --skip-any-pdfa document.pdf

# Explicitly convert a signed PDF, invalidating its digital signatures
pdftopdfa --allow-signature-invalidation document.pdf

# Convert an entire directory
pdftopdfa -r ./documents/ ./output/

# The OCR examples below use the externally managed model directories
DET_MODEL=/opt/pdftopdfa/models/PP-OCRv6_medium_det_onnx
REC_MODEL=/opt/pdftopdfa/models/PP-OCRv6_medium_rec_onnx

# OCR a German/English scan to tagged PDF/A-2a and validate it
pdftopdfa -l 2a --validate --ocr-lang de+en \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Order OCR lines by detected columns for a cleaner reading order
pdftopdfa --ocr-layout \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Add review-required ActualText to undescribed text-based image Figures
pdftopdfa -l 3a --pdfua --ocr-figure-text \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Use the same models through DirectML on Windows 11
pdftopdfa --ocr-execution-provider directml \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Automatically orient pages without deskewing them
pdftopdfa --rotate-pages \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Deskew pages without changing their 90-degree orientation
pdftopdfa --deskew \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Deskew and orient pages without converting the result to PDF/A
# (creates document_processed.pdf)
pdftopdfa --no-pdfa --deskew --rotate-pages \
  --ocr-detection-model-dir "$DET_MODEL" \
  --ocr-recognition-model-dir "$REC_MODEL" \
  document.pdf

# Preserve known proprietary stamps as PDF Stamp annotations
pdftopdfa --preserve-stamps document.pdf

The OCR examples above use POSIX shell syntax. For DirectML in PowerShell on Windows 11, for example:

$DET_MODEL = "C:\models\PP-OCRv6_medium_det_onnx"
$REC_MODEL = "C:\models\PP-OCRv6_medium_rec_onnx"
pdftopdfa --ocr-execution-provider directml `
  --ocr-detection-model-dir "$DET_MODEL" `
  --ocr-recognition-model-dir "$REC_MODEL" document.pdf

from pathlib import Path
from pdftopdfa import convert_to_pdfa

result = convert_to_pdfa(
    input_path=Path("input.pdf"),
    output_path=Path("output.pdf"),
    level="2b",
)

accessible_result = convert_to_pdfa(
    input_path=Path("input.pdf"),
    output_path=Path("accessible.pdf"),
    level="2a",
    pdfua=True,
    document_title="Annual report",
    document_language="en-GB",
)

ocr_result = convert_to_pdfa(
    input_path=Path("scan.pdf"),
    output_path=Path("scan_pdfa.pdf"),
    level="2b",
    ocr_languages=["de", "en"],
    ocr_detection_model_dir=Path("/opt/pdftopdfa/models/PP-OCRv6_medium_det_onnx"),
    ocr_recognition_model_dir=Path("/opt/pdftopdfa/models/PP-OCRv6_medium_rec_onnx"),
    ocr_execution_provider="cpu",
)

Supplying both model directories enables OCR in convert_to_pdfa(), convert_files(), and convert_directory(). Supplying only one directory, or requesting OCR through ocr_languages, ocr_force, ocr_deskew, ocr_rotate_pages, ocr_layout=True, ocr_figure_text=True, or a non-CPU execution provider without both directories fails configuration validation. convert_to_pdfa() handles this through the original-input fallback; batch-wide configuration checks still reject the invocation. Set ocr_execution_provider="directml" to use the supported DirectML configuration on Windows 11, or ocr_execution_provider="directml:1" to pass a specific raw DXGI adapter index.

Set pdfa=False to apply only the requested OCR processing. This skips font embedding, PDF/A sanitization, metadata synchronization, color-profile embedding, and PDF/A validation. The result is not validated or guaranteed to be PDF/A compliant.

See the Usage Guide for the full CLI reference, conversion API documentation, and examples. The OCR guide covers image, table, and reusable OCRSession APIs.

Limitations

• No PDF/A-1 support -- only PDF/A-2 and PDF/A-3 levels are supported
• Automatic level A semantics -- generated tags infer headings, paragraphs, lists, conservative tables, figures, artifacts, links, forms, and reading order from text styles, geometry, direct painting provenance, and, for scans, the internal OCR engine's line and layout data. Ambiguous content falls back to conservative paragraph or division structure. Trustworthy existing Alt, marked-content ActualText, and textual Captions are retained or propagated. With explicit --ocr-figure-text opt-in, sufficiently confident text from otherwise undescribed direct image Figures is written as review-required ActualText; this is replacement text, not an authoritative visual description. Direct image Figures rejected by OCR are marked as review-required Layout artifacts. Ineligible images and undescribed Formula elements remain reported for manual review. Pages with unclassified vector painting are likewise reported because decorative rules and meaningful vector diagrams cannot always be distinguished automatically. Full-page OCR scans that may contain unrecognized non-text visuals, Link annotations that cannot be associated with one content owner, and form fields without a trustworthy tooltip or field name are retained and reported for the same reason. Review automatically inferred semantics for accessibility-critical publications. PDF/A level A does not by itself imply PDF/UA conformance; request the additional PDF/UA-1 requirements with --pdfua or pdfua=True. A review_required result means both machine profiles passed but one or more semantics still require an author decision; it is not a certification. veraPDF cannot judge whether content order, descriptions, labels, language, contrast, color use, or media alternatives are meaningful. PDF/UA-2 generation is not implemented: it requires a separate PDF 2.0 and PDF/A-4 output track. The low-level validator API can inspect an existing file with the ua2 profile.
• Encrypted PDFs -- copied unchanged with a warning, including PDFs with an empty user password. OCR and validation are skipped. This also applies when PDF/UA or --validate is requested, without requiring --publish-noncompliant. The result reports success=True, skipped=True, and no produced PDF/A target; a requested PDF/UA target reports pdfua_status="not_produced". With an automatically generated output name, the unchanged copy still receives the _pdfa.pdf suffix; it is not a converted PDF/A file.
• Digitally signed PDFs -- signed PDFs are ordinarily copied unchanged because conversion would invalidate their signatures, including when --validate or --pdfua is requested. If copying succeeds, the result reports success=True, published=True, skipped=True, and target_produced=False; a requested PDF/UA target reports pdfua_status="not_produced". The unchanged copy is not a converted or validated target. Use --allow-signature-invalidation only when an unsigned archival copy is intentional.
• Font replacement -- fonts without a suitable metrically compatible replacement produce a warning; the resulting file may not be fully compliant
• Non-embedded CIDFonts (Identity encoding) -- content streams reference glyph IDs of the original font; after replacement with a substitute font the same glyph IDs point to different or missing glyphs, so the affected text may render incorrectly or invisibly. Text extraction and copy/paste stay correct because the original ToUnicode mapping is preserved. A warning is emitted for each replaced CIDFont

Font Sourcing

• On Windows, pdftopdfa may automatically embed a conservative fixed allowlist of local fonts from %WINDIR%\Fonts.
• A Windows system font is only used when the installed file lives under %WINDIR%\Fonts, its actual PostScript name is allowlisted, and its OpenType fsType permits outline embedding.
• On macOS and Linux, system fonts are never auto-embedded; bundled replacement fonts are used instead.
• fsType checks are a technical safeguard only and do not replace the font vendor's EULA or other license terms.
• For auditable deployments, keep the allowlist tied to reviewed target systems or golden images.

Development

python -m pip install -e ".[dev,ocr]"

Running Tests

python -m pytest

The test suite covers fonts, color profiles, metadata, sanitization, OCR, and end-to-end conversion.

CI additionally installs the checksum-pinned veraPDF 1.30.2 distribution and runs a real dual PDF/A/PDF/UA-1 conversion gate rather than mocking the validator boundary.

The real-model semantic OCR tests are opt-in and never modify the configured model directories. Set PDFTOPDFA_TEST_OCR_DETECTION_MODEL_DIR and PDFTOPDFA_TEST_OCR_RECOGNITION_MODEL_DIR, install veraPDF, then run:

python -m pytest tests/test_semantic_ocr_e2e.py

With the upstream veraPDF-corpus-staging checkout present at the repository root, the corpus runner converts and validates every fixture against every supported PDF/A level and writes a detailed report. It also records the exact corpus hashes and execution environment in run_metadata.json; set PDFTOPDFA_CORPUS_WORKERS to override the default worker count capped at eight:

python run_corpus_test.py

Code Quality

ruff check src/ tests/   # Linting
ruff format src/ tests/  # Formatting

Documentation

Additional documentation is available in the docs/ folder:

• Usage Guide (CLI & Python API)
• OCR Guide
• PDF/A-2/3 rules reference (veraPDF)

Contributing

Contributions are welcome! Please open an issue to report bugs or suggest features, or submit a pull request.

Dependencies

Core:

• pikepdf -- PDF manipulation (based on QPDF)
• lxml -- XMP metadata processing
• fonttools -- Font analysis, subsetting, and embedding
• pdfminer.six -- read-only digital text, layout, and painting-provenance extraction
• NumPy -- Array processing for OCR and table recognition
• click -- CLI framework
• colorama -- Colored terminal output
• tqdm -- Progress bars
• PaddleOCR -- document orientation and PP-OCRv6 text recognition

Optional:

• OCRmyPDF -- PDF rasterization, text-layer generation, and page merging for optional OCR
• ONNX Runtime -- CPU or DirectML inference for Paddle models
• PaddleX -- local-model OCR and table-recognition pipelines
• pypdfium2 -- PDF page rasterizer for OCR
• veraPDF -- external application for PDF/A and PDF/UA machine validation

Acknowledgments

This project bundles the following resources:

• Liberation Fonts -- metrically compatible replacements for the PDF Standard 14 fonts (SIL OFL 1.1)
• PP-LCNet_x1_0_doc_ori -- bundled document-orientation model (Apache-2.0)
• Noto Sans CJK -- CJK font coverage (SIL OFL 1.1)
• Noto Sans Symbols 2 -- symbol font replacement (SIL OFL 1.1)
• STIX Two Math -- math font replacement (SIL OFL 1.1)
• sRGB2014.icc -- ICC sRGB profile (ICC)
• ISOcoated_v2_300_bas.icc -- ICC CMYK profile, FOGRA39 (zlib/libpng license)
• sGray -- compact grayscale ICC profile (CC0-1.0)
• Adobe cmap-resources -- CID-to-Unicode mapping data (BSD 3-Clause)

License

This project is licensed under the Mozilla Public License 2.0 or later (MPL-2.0+) -- see LICENSE for details.