Changelog

Background OCR + Image Occlusion box refinement used to swallow failures silently, which made support harder when something went wrong ("the box is in the wrong place" → no clue why). Both paths now log a one-line diagnostic to the Anki debug console with the actual error, while still keeping the same user-visible behaviour (raw box stays in place, you can drag it manually). The Tesseract subprocess now also catches "binary missing" and "60-second hang" errors instead of letting them crash the worker — same recovery path, just no more mysterious hangs.

Bridge calls between the Cardivate UI and Python now time out after 60 seconds (longer for the few long-running ones like card generation) instead of spinning forever when something goes wrong on the Python side. If the underlying handler crashes or never responds, the UI now surfaces a clear "Try restarting Anki" error instead of a stuck loading state you'd previously have to kill Anki to escape.

Background-thread tightening. The Image Occlusion box refiner now coordinates with the rest of the preview pipeline via a proper lock, so a new generation arriving mid-refine can't silently drop refined coordinates from the previous batch. The Help / feedback dialog no longer crashes Anki on the rare path where you close it while a "send feedback" request is still in flight — the callback now detects the destroyed dialog and exits cleanly instead of poking at a freed C++ widget.

Round of internal hardening: Settings now refuses to send credential-looking keys (license keys, API tokens) to the webview at all; the in-app updater only accepts download URLs on the canonical Cardivate / GitHub hosts; the PDF Q&A pane validates the persisted last-PDF path more strictly before reopening it on launch. Server side, the legacy admin URL-token bootstrap is now subject to the same brute-force rate limit as the login form. No user-facing behavior changes if you weren't doing anything weird.

You can now generate cards from images — PNG, JPG, JPEG, or WEBP — alongside PDFs and topics. Pick one or many images at once (a textbook page photo, a slide screenshot, a hand-drawn diagram), and each becomes its own card source. Auto-detect language works per-image so mixed-language batches stay accurate. Image Occlusion is supported as a card type: hide labels on the photo and quiz yourself. Smart mode is PDF-only (there's no chapter structure to outline in a loose image set).

When you generate from multiple PDFs and one of them silently produces zero cards (usually because it's a scanned PDF without OCR), you now get a clear breakdown in the success toast (e.g. "Per-PDF: English.pdf: 8, Russian.pdf: 0") plus a console hint suggesting the fix. Before, the failure was invisible and you'd just see "8 cards generated" with no clue that one PDF didn't contribute.

Added a description under Settings → Default language to clarify its new role after v2.5.0: it's the fallback for Topic mode and for PDFs that Auto-detect couldn't identify. The per-run picker in the Create pane still overrides it.

The Language picker now tells the truth. Pick 🌐 Auto-detect from PDF and each PDF's cards are generated in its own language. Pick a specific language and every card is in that language — no surprises. The separate "Auto-detect" checkbox in Settings is gone — one control, one source of truth.

Existing users: if you had auto-detect on (the default) and never manually changed your language, you'll see Auto-detect from PDF selected automatically. Your language picks are preserved if you'd customised them.

Updated the "Auto-detect PDF language for cards" description in Settings → General to reflect how it actually works now: detection runs per PDF independent of OCR settings, and a mixed-language batch produces each PDF's cards in its own language.

Removed the manual OCR language picker from the Create pane. OCR language is now always auto-detected per PDF — the manual override was redundant for 99% of cases and added a confusing fourth language picker on top of Language, Target language, and Translate to. One less option to think about.

The per-PDF language detection from v2.4.3 only fired when the "Auto-searchable PDF" setting was on. Now it runs for any PDF source whenever auto-detect is enabled, so a mixed Russian + English batch correctly generates each PDF's cards in its own language.

Mixed-language PDF batches now generate each PDF's cards in its own language. Previously a batch of two PDFs in different languages would generate all cards in the first PDF's language — so an English PDF after an Arabic PDF got Arabic-translated cards. Now each PDF is detected separately and its cards stay in its own language.

Removed a developer-only debug toggle from Settings → Advanced. The Advanced section now only shows options that are actually useful for end users.

Recent generations now live in a sticky side rail on wide windows instead of pushing the create flow down — and collapse into a single line on narrow windows so they're still one click away.

Generation is faster, the Create pane is calmer, and progress is finally visible while it works.

A small fix to how Cardivate recognises your computer.

A small fix-and-polish update.

A big upgrade across Mock Exam, Ask a PDF, the in-reviewer tutor, and the overall look-and-feel. Highlights:

Four big additions that turn Cardivate from a card generator into a study workflow.

A small but useful study-quality tweak.

Quality-of-life polish around updates and version awareness.

License recovery, a friendlier checkout, and a cleaner Help dialog.

• Smoother checkout with automatic top-up detection — buying again on the same email adds credits to your existing license instead of issuing a new one.
• Faster license activation after payment confirmation.
• Clearer error messages when a license is at its device limit.

• Image occlusion editor — generate masks from PDF diagrams; drag, resize, and re-order them before adding to your deck.
• Scanned PDF support — generate cards from image-only PDFs that have no embedded text.
• Hint badges on occlusion cards (e.g. "1/4") to help with orientation during review.

• Cloze and Basic Q/A generation from any page range, with per-card preview and inline editing before adding to a deck.
• Optional page thumbnails attached to cards.
• Multi-language output regardless of the source PDF's language.

• First public release of the Cardivate Anki add-on.
• License + credit system with a self-serve portal and a free trial.
• PDF → flashcards with cloze, basic Q/A, and image-occlusion card types.