Skip to main content
Version: V12

Release Notes 12.2.363

19 August 2026​

A lean release that punches above its weight. One new feature for anyone who works with hard-to-read text, a round of enhancements to the tools you already lean on, and a long list of fixes that clears out the everyday snags.

What's New​

  • Enhanced Small-Text Reading: A plate at distance, a street sign, a caption in the corner: small text sits below what a downscaled frame can resolve. Turn on Enhanced small-text reading in the process modal and VIDIZMO takes a second pass at full source resolution once the main job wraps. Works on images, documents, and video that isn't rotated, and adds a little processing time on high-resolution content.

  • Your AI assistant can now read scanned and image-only PDFs, so it can answer questions even when a document has no selectable text to search.

  • Automated workflows can now start on their own when a relevant event happens, instead of only when someone starts a chat.

  • You can now see the history of your automated workflow runs, including status and error details, from your portal settings.

Enhancements​

  • The Object Library got easier to keep tidy. Clear out several subjects at once with the list refilling as you go, or pause recognition on a subject so cameras stop matching it while it stays enrolled.

  • Object detection runs on a retrained model. Thin classes, plus fire hydrants, graffiti, and fallen trees, come back noticeably more accurate.

  • Folders and cases can carry a thumbnail now. Give them an image and they stop looking identical halfway down a long list.

  • Tool nodes in your agent graphs inherit the agent's AI model by default, with a per-tool override when a lighter model will do. Read Content now sets its own read depth, and Transcribe Media publishes what it transcribes.

  • Webhook deliveries hold up when part of the system drops offline. Deliveries that would have been stranded get picked up and finished.

Bug Fixes​

  • Word documents no longer get shrunk before text recognition reads them, so no more missing words.

  • Fixed the header-only text extraction on scanned and image-only PDFs! The whole page comes through now.

  • The OCR language dropdown stops hiding in the process modal, and remembers the language you picked.

  • Low-confidence license plate re-reads could be reported as confirmed plate text. Weak reads are discarded instead.

  • Object Library requests stopped failing with a server error, so enrolled subjects load and save again.

  • Content lifecycle and archival settings could show or store a different option than the one selected. The saved value now matches your selection.

  • Admin screens no longer show raw internal labels: the Document Studio sidebar and PII help text included.

  • Add New looks like the primary action again, and quits hugging the edge on narrow screens.

  • The left navigation's collapse control was an unlabelled icon that gave no clue what it did. It now carries a visible Collapse label.

  • The footer copyright line was rendering at an oversized font in production. It's back to a subtle footer size.

  • Play time was inflated by paused and idle stretches counted as playback. Reported watch time now reflects what actually played.

  • Turning off Search Engine Visibility now actually tells crawlers to stay out.

  • Chat and chat session events now file into the audit log under one name, not two.

Did you find this article helpful?