v1.22.1 v1.23.0
AIPentest/CyberStrikeAISep 20, 2026 → Sep 22, 20261 day between releases
What's New
Introduces text-layer repairs including handling large Form XObject bounding boxes and fixing right-to-left text reading order.
Breaking Changes
- Right-to-left text extraction now prioritizes visual order over logical order for visible runs, potentially altering output text order.
New Features
- Detects invisible text layers under images as a separate OCR reason
- Interpolates single-glyph gaps in ToUnicode CMaps to mark ambiguous glyphs as U+FFFD
Improvements
- Saturates large numerals in Form XObject bounding boxes to prevent empty page extraction
- Refines ligature glyph decoding for re-encoded simple fonts with stale ToUnicode CMaps
Bug Fixes
- Fixes spacing accent composition to attach accents to the correct letter instead of leaving them stray
- Fixes right-to-left text reading order for mirrored pages to match visual appearance
- Handles zero-advance dependent signs to maintain word integrity in attached scripts
Upgrade Notes
Upgrade to v1.23.0 to improve extraction of mirrored pages and invisible text layers, but verify text order handling as the logic for right-to-left text has changed to prioritize visual order. The parser now safely handles large bounding box values in Form XObjects.
The Two Releases
1.22.1 — Ligature glyph names
Sep 20, 2026· 1 month ago
Version 1.22.1 adds support for reading ligature glyph names by their components instead of dropping them, improving text extraction accuracy.
Full release notes1.23.0 — Text-layer repairs
Sep 22, 2026· 1 month ago
Version 1.23.0 focuses on text-layer repairs, including handling large Form XObject bounding boxes, fixing right-to-left text reading order, and improving ligature glyph name decoding.
Full release notes