Fixed unwanted blank lines in the Text view, including number lists and tables. Editing recognized text now keeps line breaks stable, making it easier to copy a compact list of numbers into a spreadsheet. The Markdown table preview remains available.

Screen capture now shows whether it is selecting, reading, or recognizing your screenshot. Clearer first-run guidance explains that initial text recognition can take a minute or more while macOS gets ready. OwlOCR 7.21.1 adds detailed timing diagnostics for screen capture and text recognition to help investigate slow or stalled captures. It also fixes an issue that could prevent support emails with diagnostics attached from opening. Mac launches now use the native version of OwlOCR for your processor. Both Intel and Apple silicon Macs remain supported. This update also includes the improvements from OwlOCR 7.21: - Automatically detect document languages, including Korean, with Apple Vision in the app and Finder Quick Actions. - Keep recognized text when a PDF contains a damaged or unreadable embedded text layer. - Prevent screen capture from waiting indefinitely for stalled system capture or recognition processes. - Improve capture cancellation and show recovery guidance when capture cannot finish. - Update the app, PDF processing, local OCR AI runtime, and website components.

OwlOCR 7.21 improves Korean recognition, PDF text handling, and screen capture reliability. - Automatically detect document languages, including Korean, with Apple Vision in the app and Finder Quick Actions. - Keep recognized text when a PDF contains a damaged or unreadable embedded text layer. - Prevent screen capture from waiting indefinitely for stalled system capture or recognition processes. - Improve capture cancellation and show recovery guidance when capture cannot finish. - Update the app, PDF processing, local OCR AI runtime, and website components.

Move through documents with quicker page previews, less repeated OCR work, and more reliable searchable PDF preparation. New: - In the direct-download edition, turn off Automatically check for updates in Settings > General to stop background update prompts. Your choice saves immediately, and Help > Check for Updates remains available. Mac App Store updates remain managed by the App Store. Improved: - Keep recently viewed pages ready and prepare the next preview while the app is idle. - Reuse completed OCR AI results sooner when your documents and recognition settings are unchanged. - Reduce repeated image preparation and make searchable PDF preparation more efficient. - Update local OCR AI and PDF processing components, with more reliable cancellation and cleanup. Fixed: - Keep image orientation and mirroring aligned between previews, recognized text, and barcode positions. - Match Finder Quick Action file ordering to the app, so the first scan precedes its numbered continuation pages. - Handle invalid page numbers safely when navigating documents. - Make OCR All recognize text on pages previously scanned only for barcodes while preserving completed text and your edits.

Searchable PDF export now handles Finder tags and other file metadata changes reliably. Fixed: - Prevented a false “file changed while preparing the PDF” error when your image or PDF contents are unchanged. - Kept source checks responsive to cancellation while confirming that recognized text still matches the original file.

Work more smoothly with long OCR results, searchable PDFs, and local OCR AI. Improved: - Convert long Markdown and HTML results to plain text faster while preserving line breaks, tables, and image descriptions. - Prepare searchable PDFs more efficiently while preserving recognized text and the original page appearance. - Check whether local OCR AI is ready more quickly, with less background disk activity. - Keep keyboard shortcuts and menu bar controls steady when you change unrelated preferences. Fixed: - Protect saved preferences from incomplete writes when saving is interrupted.

Export searchable PDFs faster by reusing OCR you have already completed. Your corrected text stays intact. Improved: - Reuse completed OCR when your document and recognition settings are unchanged, and recognize only pages that need it. - Preserve your text corrections and automatically detected page rotations during export. - Keep completed OCR results if another document fails, reducing repeated work when you retry. - Reduce repeated processing while preparing searchable PDFs and displaying long Markdown output. Fixed: - Preserve literal dollar signs and code snippets in Markdown previews. - Reuse successful recognition of blank pages instead of running OCR again.

Need OCR text with no empty rows? Settings → Fixes now includes Remove all blank lines, alongside Preserve and Collapse repeated blank lines. New: - Remove every empty or whitespace-only line while keeping ordinary line breaks between recognized rows. - Use the same cleanup in Apple Vision OCR, OCR AI, the command line tool, and Finder Quick Actions. Improved: - Preserve existing blank-line preferences during upgrades and normalize conflicting settings safely. - Handle LF, CRLF, CR, and Unicode line separators consistently. - Expand regression coverage for the app, native helper, CLI, settings, and Finder extensions.

Searchable PDF export now finishes reliably even when optional layout analysis stalls. Fixed: - Keep Export Single Searchable PDF available after OCR completes when optional LiteParse processing times out or fails. - Prevent late progress events from replacing finalized OCR results or leaving the interface busy. Improved: - Run LiteParse in an isolated, cancellable worker with strict deadlines and completed-OCR fallback. - Keep OCR and export state consistent through cancellation, timeout, failure, and worker shutdown. - Stage sandbox PDFs with abortable, permission-restricted temporary files and clean them after completion or crashes. - Update Electron, LiteParse, the local llama.cpp OCR AI runtime, Next.js, Wrangler, and related dependencies. - Expand regression coverage for finalization, worker lifecycle, cancellation, sandbox staging, runtime pins, and packaging.

This maintenance update makes OwlOCR safer and more reliable while preserving its existing workflows. Fixed: - Render inline code and image placeholders correctly when they appear inside Markdown link labels. - Prevent internal Markdown placeholder markers from appearing in previews or plain-text cleanup. Improved: - Validate saved settings field by field while preserving existing defaults, migrations, and entitlement safeguards. - Check native OCR responses, OCR cache entries, and renderer-to-main messages before using their data. - Strengthen file, URL, diagnostics, and export boundaries against malformed input. - Expand regression coverage across settings, OCR payloads, app communication, caching, Markdown rendering, and packaged dependencies.

Searchable PDF exports now keep every recognized table word in its detected position while removing only unwanted Markdown or HTML table formatting. Fixed: - Preserve all OCR AI table text as searchable, line- and cell-positioned content when “Preserve OCR AI table formatting” is turned off. - Flatten Markdown and HTML table structure without deleting the words inside each cell. - Prevent exports from retaining only the first heading or text block on table-heavy pages. - Keep right-margin references in separate boxes and preserve the detected horizontal width of each line. - Handle escaped and encoded pipes, inline code, entities, comments, quoted attributes, and multiline table markup without losing searchable text. LightOnOCR 2 remains the recommended default, and OvisOCR2 remains available as an optional experimental model.

OCR AI now gives you a second local model choice: OvisOCR2 is available as an optional experimental model for structured pages and may process some documents faster with less memory. New: - Choose OvisOCR2 or the recommended LightOnOCR 2 from Settings. - Keep both models installed and switch without deleting the other model’s files. - Use the same saved model choice from OwlOCR’s command line tool. Improved: - Ovis downloads a pinned, checksum-verified model conversion and runs entirely on your Mac. - Model-specific prompts, inference profiles, caches, and output cleanup keep results isolated when you switch models. - Ovis removes unsupported image-crop placeholders and trims pathological repeated endings from extracted text. LightOnOCR 2 remains the recommended default and has the broadest validation for degraded archival, multilingual, and scientific documents. OvisOCR2 currently uses Balanced Q4 and processes one page at a time.

Searchable PDFs now keep OCR AI text aligned with the words on the page—even across paragraphs, columns, and edited text. Improved: - Position semantic OCR AI paragraphs on their matching physical rows without compressing them into one hidden line. - Preserve whole words, headings, footers, and column order when rebuilding searchable layers. - Refresh stale searchable text only when replacement geometry covers the complete page text. Fixed: - Exclude divider-less Markdown tables, including rows with blank cells, when table embedding is disabled. - Keep pipe-delimited math, shell commands, and code examples searchable. - Consolidate linked wrapped-line controls into a single setting under Fixes.

OwlOCR 7.13.3 refreshes local OCR runtimes and makes Mac releases leaner and safer. Improved: - Upgrade LiteParse to 2.11.1 and the bundled llama.cpp inference runtime to b10310. - Exclude unused Linux and Windows OCR components from Mac downloads. - Validate update artifacts, blockmaps, sizes, and checksums before publishing. - Include the 7.13.2 searchable PDF text-placement, table-filtering, and compact scan-image fixes.

OwlOCR 7.13.1 improves searchable PDF accuracy and OCR AI table handling. Improved: - Keep searchable OCR text aligned with each detected line on the page. - Add paragraph semantics for compatible PDF readers without moving the text layer. - Clarify how wrapped-line handling behaves in OCR text and searchable PDFs. Fixed: - Omit both Markdown and HTML tables when OCR AI table embedding is disabled. - Preserve literal table examples inside code and comments. - Record wrap and text-cleanup settings in diagnostic logs for easier troubleshooting.

OCR AI text now lands where it belongs in searchable PDFs, while corrected scans stay dramatically closer to their original file size. Improved: - Uses an on-device Apple Vision layout pass to align OCR AI text with the page's real lines and columns. - Remembers page layout in the private OCR AI cache for fast, accurately positioned repeat recognition. - Preserves original compressed scan imagery whenever safe and overlays only changed pixels. Fixed: - Avoids massive file-size growth on common compressed scans. - Keeps multi-column text, page rotations, vector details, and corrected searchable text aligned. - Restores your previous page text and layout when OCR is cancelled. - Prevents cache clearing or late background work from overwriting newer results.

Huge scanned PDFs now use far less memory while keeping every page visually intact and searchable. Improved: - Keeps memory bounded while writing long searchable PDFs, including Separate PDF exports. - Renders OCR pages at the requested pixel resolution, improving speed without sacrificing PDF accuracy. - Inspects embedded PDF text in small batches so very large documents remain responsive. Fixed: - Replaces stale OCR text on scanned pages that also contain small vector or inline elements. - Keeps page geometry, metadata, and visible content intact when rebuilding searchable layers. - Pauses, resumes, and cancels every export worker together, with reliable cleanup even during immediate cancellation.

OwlOCR 7.11.3 speeds up multipage recognition and refreshes the local OCR AI runtime. Improved: - Uses adaptive image resolution, on-device result caching, and safe PDF and blank-page fast paths to avoid unnecessary AI work. - Processes PDF and image pages concurrently while preserving document order. - Shows a rough time-remaining estimate during multipage OCR and PDF export. - Simplifies OCR AI settings, adds optional small-image upscaling, and includes a one-click reset to recommended defaults. - Lets OCR AI handle up to two pages at once with bounded memory use. - Lets the inference runtime select supported Metal acceleration automatically. - Prevents overlapping work from starting duplicate local AI servers or overwriting newer settings. - Updates the bundled local inference engine for current model and runtime compatibility.

Shutting down or restarting your Mac is smooth again—OwlOCR now closes promptly instead of holding up the process or requiring you to quit it manually. Fixed: - Lets macOS close OwlOCR immediately during shutdown and restart. - Keeps normal Quit behavior tidy by stopping active OCR and local AI processes before exiting. - Delivers the same reliable shutdown behavior on Intel and Apple silicon Macs.

OwlOCR 7.11.1 makes toolbar tooltips less intrusive. Improved: - Hides a tool's tooltip as soon as you begin a screen selection. - Adds a setting under General > Toolbar to disable tooltips across the main toolbar. - Keeps toolbar controls labeled for VoiceOver when visual tooltips are turned off.

OwlOCR 7.11.0 can automatically correct turned pages. New: - Adds an opt-in orientation check for sideways and upside-down pages. - Keeps previews, OCR text boxes, and searchable PDF exports aligned with the detected rotation. - Supports both Apple Vision and OCR AI while preserving manual page rotations.

OwlOCR 7.10.2 improves OCR reliability. Fixed: - Prevents successful OCR from failing when macOS emits an unexpected native diagnostic message. - Makes communication with the OCR helper more resilient during single-page and multipage recognition.

OwlOCR 7.10.1 fixes Pro access for command-line workflows. Fixed: - The command-line tool now recognizes active OwlOCR Pro licenses from the direct-download app. - CLI entitlement checks use the correct settings when both the direct-download and Mac App Store editions have been installed.

OCR AI is now more reliable on long and dense documents, with fewer repeated text loops and an updated local runtime. Improved: - OCR AI follows the recommended LightOnOCR generation and image settings. - Repetition controls prevent documents from getting stuck repeating the same text. - The bundled local AI runtime has been updated for improved compatibility. - Numbered scanner files open in a more natural page order.

OwlOCR 7.9.0 makes large searchable PDF exports faster and easier to follow. Improved: - Searchable PDF writing is much faster when OCR text contains narrow text boxes. - Separate searchable PDF exports show page-level progress with the current filename and document count. - PDF exports can be cancelled while files are being written. - Account numbers and similar text remain contiguous in the searchable PDF layer.