Skip to content

Changelog

The changelog is a record of changes to the DeepVA software. Choose between a high-level Overview and the Detailed release notes for every DeepVA component.

A summary of the most important changes in DeepVA, newest first. Switch to Detailed to see every change for each component.

2026-10-15

  • Auto merge profiles can be managed in the Preferences and through the API.
  • More flexible job results exports and mining module settings.
  • Imports of up to 10 GiB per file, plus improvements to transcript editing.
  • Fixes for on-premise installations without internet access, keyboard shortcuts and dictionaries.

2026-10-07

  • Lower Third Recognition detects names more accurately and now also extracts the description below a name, such as a job title or role.
  • Visual Understanding uses more accurate shot detection by default.

2026-09-30

  • DeepVA 2.0: a completely redesigned user interface, with light, dark and brand themes, full keyboard navigation and much faster loading.
  • The interface is now also available in Czech, French and Spanish.
  • Professional transcript editing, including undo/redo, find & replace, quality review, paragraph approval, a speakers overview and CSV export.
  • Post-workflows guide you through their required parameters and mining modules.
  • Move and rename files and folders in Storage, and a redesigned Imports / Exports page.
  • New Qwen 3.6 and Qwen 3.8 models for Visual Understanding in on-premise installations.

2026-09-22

  • Post-workflows can define which mining modules they require.
  • Improvements to class custom fields and dataset imports.

2026-09-08

  • Dataset imports, with a combined Imports / Exports center.
  • Stability fixes on the job results page.

2026-09-03

  • Datasets can be imported through the API.

2026-08-13

  • Redesigned job results page for all mining modules, which now handles very large result sets smoothly.
  • Reworked class auto merge, with matching on custom fields, a redesigned dialog and clear merge reports.
  • Many display fixes for bounding boxes, timecodes, right-to-left text and thumbnails.

2026-08-04

  • New interactive timeline, which shows the results of all mining modules together, with zoom, search and a playhead synced to the video.
  • Redesigned jobs page, job wizard and mining module settings, including selection of multiple models.
  • Faster loading of thumbnails and pages.

2026-07-24

  • Post-workflows are synchronized automatically, and thumbnails are delivered through secure links.

2026-07-20

  • An optional advanced settings step in the job wizard.
  • A face sharpness setting for Face Recognition.
  • On-premise login improvements.

2026-07-08

  • The jobs and job results pages show the state of each post-workflow.
  • Export artifacts are downloaded directly in the browser.

2026-07-01

  • Post-workflows and their status are available through the API.
  • Fixed downloads of professional video formats such as DNxHD.

2026-06-25

  • Face Recognition can skip blurry faces.

2026-06-23 – 2026-06-24

  • Much faster Face Recognition when indexing many faces.
  • Fixes for Speech Recognition paragraph formatting, bounding box alignment and custom Face Recognition models.

2026-06-17 – 2026-06-22

  • Post-workflows are shown in the web application, including their results on the job results page.
  • Bounding boxes for Face Recognition results.
  • Larger Qwen 3.5 models for Visual Understanding in on-premise installations.
  • Better form validation, a security fix and many smaller fixes for jobs, datasets and classes.

2026-05-29

  • Post-workflows run automatically after a job finishes.
  • Consent management for modules that require you to accept terms of service.
  • Webhook events for when a job starts, completes or fails.
  • New Gemma 4 model for Visual Understanding, and dictionaries can be updated in place.

2026-04-20 – 2026-04-28

  • Consent management in the job wizard, the training wizard and for API keys.
  • Select and delete several items at once, and model selection for Visual Understanding.
  • The Transcript Editor is now available in on-premise installations.
  • Faster video playback, higher API rate limits and many usability fixes.

2026-04-13 – 2026-04-14

  • Improved translation of Speech Recognition results, with a new translation engine.
  • New Qwen 3.5 models and support for memory-efficient quantized models in Visual Understanding.
  • Black frame detection in Color Detection, and progress reporting in Speech Recognition.

2026-02-24

  • Fixed seeking in videos stored in S3 storage.

2026-02-05

  • Visual Understanding can split videos into shots of a fixed length, has new Qwen3-VL models (Instruct and Thinking) and handles long videos reliably.
  • A choice of engine for Speech Recognition, and faster Text Recognition.
  • Results of completed modules are shown while a job is still running.

2025-12-16 – 2025-12-18

  • More accurate and faster Speech Recognition models for on-premise installations.
  • Support for MXF files without duration information, and more reliable sessions.

2025-11-10

  • Security review of the web application, plus fixes for dictionaries and index collections.

2025-08-20

  • Visual Understanding can return structured answers that follow an output schema you define.

2025-07-31

  • A new, more secure login and account system.

2025-07-18 – 2025-07-21

  • Support for multi-tenant and on-premise deployments.
  • Visual Understanding results are shown as formatted text or structured data, with new settings for shot detection and model selection.
  • New settings for Text Recognition.

2025-06-18

  • GPU-accelerated Text Recognition, and higher processing capacity for Visual Understanding and Text Recognition.

These release notes list the new features, improvements and fixes in DeepVA. DeepVA is made up of several components, and each has its own version number:

  • Frontend: the DeepVA web application
  • API: the public DeepVA REST API
  • Manager: job orchestration, mining module configuration, datasets, transcripts and workflows
  • DeepVA Worker: runs the mining modules (AI services) such as Speech Recognition, Face Recognition and Text Recognition
  • Visual Understanding Worker: runs the Visual Understanding mining module Releases are listed newest first.

2026-10-15

Frontend 2.1.0

  • New: Auto merge profiles. You can now create, edit and delete auto merge profiles in a new Preferences section. Every Preferences section now has its own link.
  • New: Choose which mining module and post-workflow results go into a job results JSON export.
  • Improved: Mining module settings can now offer multi-select dropdowns and options that change with other settings, for example settings that depend on the selected model.
  • Improved: Restarting a job no longer asks for hidden settings of modules you did not open.
  • Fixed: Fonts and icons now display correctly in on-premise installations without internet access.
  • Fixed: Zoom-out on German keyboards and ⌘+click on macOS work again. Keyboard shortcuts no longer interfere with buttons, menus and text fields.
  • Fixed: An edited dictionary now shows and downloads its latest content.

API 1.73.0

  • New: You can create, update and delete auto merge profiles through the API.
  • Improved: The maximum file size for imports was raised from 5 GiB to 10 GiB.
  • Fixed: A permission error on existing media storage after an update.

Manager 1.33.2

  • Improved: Saving a transcript version can now remove merged paragraphs. Edited paragraphs keep their linked speaker identities, and unchanged paragraphs keep their approval state.
  • Fixed: Model selection for Face Recognition.

2026-10-07

Manager 1.33.1

  • Improved: Lower Third Recognition has new options for better name detection, subtitle handling and face feedback.
  • Improved: Visual Understanding now uses content-based shot detection by default, which gives more accurate shot boundaries.

DeepVA Worker 3.3.1

  • Improved: Lower Third Recognition detects names more accurately.
  • New: Lower Third Recognition now also extracts the description shown below a name, such as a job title or role.

2026-09-30

Frontend 2.0.0 – Major release

DeepVA has a completely redesigned user interface.

  • New look and feel. The whole application has a new, consistent design. This covers the jobs, job results, knowledge graph, preferences, API keys, index collections, dictionaries, AI models and transcripts pages.
  • New: Themes. Choose a light, dark or system theme, or the DeepVA brand theme, and adjust the text size.
  • New: Interface languages. The interface is now available in Czech, French and Spanish. The language picker shows each language's name both in that language and in the current language.
  • New: Command palette and full keyboard navigation across the product.
  • Faster: The application loads much faster, because the initial download is now less than half its previous size.
  • New: Professional transcript editing in both the Transcript Editor and the embeddable editor:
    • Edit words with undo/redo; every save creates a new version
    • Find & replace
    • Quality review and navigation between approvals
    • Three-state approval for each paragraph
    • Playback that follows along, with keyboard shortcuts
    • A speakers overview
    • A new export dialog that adds CSV export
  • New: Post-workflow parameters. The job wizard fills in each post-workflow's parameters for you, with hints and the right input types, and asks for any that are required. Post-workflows that need a mining module you have not selected are greyed out, and the wizard shows which module is missing.
  • New: Terms of service consent. The job wizard asks you to accept the terms of service of any selected module that requires it and remembers your consent for each module.
  • New: Storage management. Move, rename and select files and folders in Storage.
  • Improved: A redesigned Imports / Exports page.
  • Improved: The job results player can be docked. Selecting an item on the timeline highlights the matching result card.
  • Improved: Redesigned login pages, a widescreen job wizard with keyboard shortcuts, a clearer source step, card-based advanced settings and a post-workflow catalog with formatted descriptions.
  • Improved: Layouts now work on narrow and portrait-format windows.
  • Improved: Lists remember their page, page size, filters and sort order when you navigate back.
  • Improved: For on-premise installations: optional built-in monitoring and log dashboards.
  • Changed: Once a class custom field exists, its name and type can no longer be changed.

API 1.71.0

  • New: CSV export for transcripts.
  • New: You can move and rename files and folders in storage. Their IDs and links stay the same.
  • New: For on-premise installations: optional monitoring and log dashboards.
  • Fixed: Edited dictionaries no longer show their old content because of browser caching.
  • Fixed: Dataset imports no longer fail on classes that have an empty reference. Classes whose reference already exists are skipped and listed in the import result.

Manager 1.33.0

  • New: Post-workflows can declare parameters, and jobs missing a required parameter are rejected with a clear error.
  • New: Qwen 3.6 (27B and 35B-A3B) and Qwen 3.8 (27B) models for Visual Understanding in on-premise installations.
  • Fixed: You can now sort a dataset's classes by number of samples and by number of sources, together or with other sort fields.
  • Fixed: Once a class custom field exists, its name and type are protected from accidental changes.

2026-09-22

Manager 1.32.0

  • New: Post-workflows can declare which mining modules they require, and a job that lacks a required module is rejected with a clear error.
  • New: Support for converting the type of class custom fields.
  • Fixed: A job that references an unknown or removed post-workflow now returns a clear validation error.
  • Fixed: Creating a class with an empty reference now generates a unique reference automatically.

2026-09-08

Frontend 1.71.0

  • New: Dataset imports and a combined Imports / Exports center.
  • Fixed: Clicking through result cards no longer freezes the job results page.
  • Fixed: Post-workflows without parameters are displayed again.
  • Fixed: Results that have only a start or only an end time keep their timecode and their place on the timeline.

2026-09-03

API 1.70.0

  • New: Dataset import. Import datasets into DeepVA. Imported samples are stored in the matching class folders.

2026-08-13

Frontend 1.70.0

  • Improved: Redesigned job results page. The page and the results views of all mining modules have a consistent new design with result cards, confidence badges, frame navigation and empty states.
  • New: Job artifacts are shown on the job results page for every module that produces them.
  • New: Loading placeholders while results and the timeline load.
  • Improved: The job results page now handles very large result sets without freezing.
  • Improved: Redesigned auto merge dialog. It has a searchable profile list, a thumbnail grid for classes and datasets, clearly worded merge modes, custom fields in the match criteria and full keyboard navigation.
  • Fixed: Many display issues on the job results page:
    • Bounding boxes on letterboxed images
    • Right-to-left text
    • Timecodes and rounding of confidence values
    • Duplicate timeline segments
    • Playback of storage videos while a job is running
  • Fixed: Class thumbnails load reliably and lazily, so large classes open faster.
  • Fixed: The custom fields of a class update correctly after an auto merge, merge, move, copy or delete.

Manager 1.31.0

  • Improved: Reworked class auto merge.
    • Match classes on custom fields
    • Manage profiles through the API
    • Merge modes are respected
    • Results report which classes were merged and which were skipped for each destination
    • Custom field values are kept when classes are merged
  • Fixed: A post-workflow that cannot be started is now recorded as failed, with a clear error message.

2026-08-04

Frontend 1.69.0

  • New: Interactive timeline on the job results page:
    • A draggable playhead synced to the video
    • Zoom and pan with an overview minimap
    • Show, hide, isolate and search lanes
    • Scrolling for many lanes
  • New: Unified timeline that shows the results of all completed mining modules at once. Module groups can be collapsed, each module has its own colour, and you can filter rows across modules.
  • New: Redesigned mining module settings, including ordered selection of multiple models.
  • Improved: Redesigned jobs page and a step-by-step job wizard with a review step.
  • Improved: Text Recognition and Advanced Speech Recognition results are now placed at the correct time on the timeline.
  • Improved: Face Recognition thumbnails and Advanced Diversity Analysis details now appear in timeline tooltips.
  • Improved: Faster loading thanks to more efficient network handling.

API 1.69.0

  • Improved: Thumbnails are now created when files are uploaded and delivered securely, so they load faster.

Manager 1.30.2

  • Improved: The post-workflow catalog now picks up renamed, updated and removed workflows automatically. Removed workflows are hidden from the list.

2026-07-24

API 1.68.0

  • Improved: Thumbnails are delivered through secure, signed links. Missing thumbnails are regenerated automatically.

Manager 1.30.0

  • New: The available post-workflows are synchronized automatically.

2026-07-20

Frontend 1.68.0

  • New: An optional advanced settings step in the job wizard.
  • Improved: A redesigned screen after login.
  • Fixed: On-premise login no longer shows a misleading "Forbidden" error message.

Manager 1.29.0

  • New: Face Recognition has a new face sharpness setting.
  • Improved: Post-workflows start more reliably.

2026-07-08

Frontend 1.67.2

  • New: The jobs and job results pages show the state of each post-workflow.
  • Improved: Export artifacts are downloaded directly by the browser.
  • Fixed: The sample counter on the class page no longer shows "0 of 0" when the page first loads.

2026-07-07

API 1.67.1

  • Improved: Export artifacts can be downloaded directly in the browser.

2026-07-01

API 1.67.0

  • New: List the available post-workflows and see their execution status.
  • Fixed: Downloads of some professional video formats (such as DNxHD) no longer fail.

2026-06-25

DeepVA Worker 3.1.3

  • New: Face Recognition has a face sharpness setting for skipping blurry faces.

2026-06-24

Frontend 1.67.1

  • Fixed: Bounding boxes now line up with the video for every aspect ratio.
  • Fixed: "Go to Training Source" now works for Face Recognition with custom models.

2026-06-23

DeepVA Worker 3.1.2 / 2.3.2

  • Improved: Face Recognition jobs that index many faces are much faster.
  • Fixed: Speech Recognition no longer hangs when paragraph formatting is enabled.
  • Fixed: "Go to Training Source" now works for Face Recognition with custom models.

2026-06-22

Manager 1.28.1

  • Fixed: Dataset cloning and post-training evaluation no longer get stuck as pending.

2026-06-18

Frontend 1.67.0 / Manager 1.28.0

  • Fixed: Editing a class no longer sets an expiration date when the field is left empty.

2026-06-17

Frontend 1.65.0 / 1.66.0

  • New: Post-workflows. Jobs that have post-workflows are marked with an icon, the post-workflows are listed in the job summary, and their results are shown on the job results page.
  • New: Face Recognition results show bounding boxes, and they render faster.
  • New: After you log in, links you opened before logging in take you to the page you requested.
  • Improved: Better form validation for AI models, datasets, storage and the job wizard.
  • Improved: Custom Visual Understanding settings are kept when you re-run a job.
  • Improved: Settings that contain nested objects are displayed correctly.
  • Fixed: Restarting a job with a custom Face Recognition model no longer shows a "Model does not exist" error.
  • Fixed: Job duration is shown again on the jobs page and in the job details.
  • Fixed: "Reset" in sort and filter dialogs now restores the initial values, including dates.
  • Fixed: Security fix for timeline tooltips.
  • Fixed: Many smaller layout and display issues, including in Firefox and in on-premise installations.

Manager 1.27.2

  • Improved: Finer control over Visual Understanding output with a lower minimum temperature.
  • New: Larger Qwen 3.5 models for Visual Understanding in on-premise installations.

2026-05-29

API 1.66.0

  • New: Post-workflows. Attach post-workflows to a job, and they run automatically after the job finishes. Their results can be retrieved through the API.
  • New: Consent management. You can see which modules require you to accept terms of service, and you can give consent when you start a job through the API.
  • New: You can update the content of an existing dictionary.
  • Fixed: Files with special characters in their names can now be read from DeepVA Storage.

Manager 1.27.0

  • New: Post-workflows that run automatically after a job finishes.
  • New: Modules can require you to accept terms of service, with a link to the terms.
  • New: Webhook events for when a job starts, completes or fails.
  • New: For on-premise installations: you can set a custom job priority through the API.
  • New: Gemma 4 model for Visual Understanding.

2026-04-28

Frontend 1.64.5

  • Improved: Better validation and error messages when you add a storage.
  • Fixed: The arrows on slide buttons are visible again.

API 1.65.1

  • New: A history of consents is now stored for compliance purposes.

2026-04-22

Frontend 1.64.2 – 1.64.4

  • New: The Transcript Editor is now available in on-premise installations.
  • Fixed: Login, logout and session handling in on-premise installations.
  • Fixed: Several display issues on the datasets pages.

2026-04-20

Frontend 1.64.0

  • New: Select and delete several storage files, transcripts or jobs at once.
  • New: Consent management in the job wizard, the training wizard and on the API keys page.
  • New: Model selection for Visual Understanding.
  • Improved: Videos start playing right away instead of after a full download.
  • Improved: The knowledge graph uses infinite scrolling for better performance.
  • Improved: Busy servers are handled gracefully: the app retries requests automatically.
  • Improved: Password reset now goes through the central account service.
  • Fixed: .mxf files are now accepted by drag and drop in the storage uploader.
  • Fixed: Job list filters are no longer reset while jobs are running.
  • Fixed: The storage search filters on the correct field, and the page count after filtering is correct.
  • Fixed: Videos from public URLs play on the job results page, and results update correctly when you switch between several public images.
  • Fixed: Restarting a job with a different module now uses the correct default values.
  • Fixed: Several fixes to the Transcript Editor, to tooltips and to dropdowns inside dialogs.

API 1.65.0

  • New: API keys support consent to terms of service.
  • Improved: Higher API rate limits.
  • Improved: Upgraded database for better performance and reliability.
  • Fixed: An expired login now returns the correct "unauthorized" response.
  • Fixed: Missing titles for Face Recognition and Text Recognition on new installations.

2026-04-14

Translation Service 1.0.0

  • Fixed: Translation is more stable when many requests run at the same time.

2026-04-13

DeepVA Worker 2.2.0

  • New: Improved translation of Speech Recognition results, with a new translation engine and standard language codes. Only supported language pairs are offered.
  • New: Speech Recognition reports its progress.
  • New: Color Detection can now detect black frames.
  • Improved: More stable processing of mining jobs.

Visual Understanding Worker 0.3.1

  • New: Qwen 3.5 models for Visual Understanding.
  • New: Support for quantized, memory-efficient Visual Understanding models.
  • Fixed: Internal reasoning text no longer appears in Visual Understanding answers.
  • Fixed: An error when metadata extraction is enabled.

Manager 1.26.0

  • Improved: Translation languages use standard ISO 639-1 language codes.
  • New: A black frame detection option for Color Detection.
  • Fixed: Paragraphs could be missing when you created a transcript from an Advanced Speech Recognition job.

2026-02-24

API 1.64.3

  • Fixed: Seeking in videos stored in S3 storage works again.

2026-02-05

Frontend 1.63.8 – 1.63.12

  • Improved: Results of completed modules are shown while other modules of the job are still running.
  • Improved: Text fields in the mining module wizard no longer have an unnecessary length limit. Visual Understanding prompts have a sensible maximum length.
  • Fixed: "Go to training source images" in Face Recognition results.

API 1.64.0

  • Improved: Smooth video seeking and playback for logged-in users.

Manager 1.25.0

  • New: More settings for Visual Understanding, including fixed-length shots.
  • New: A choice of engine for Speech Recognition.
  • Improved: Faster Text Recognition processing.

Visual Understanding Worker 0.2.0 / 0.2.1

  • New: Split videos into shots of a fixed length, optionally combined with shot detection.
  • New: Qwen3-VL models for Visual Understanding in several sizes, as Instruct and Thinking (reasoning) variants.
  • New: More settings, with content-based shot detection as the default.
  • Fixed: Long videos no longer run out of memory, with or without shot detection.

2025-12-18

DeepVA Worker 2.1.0 (on-premise)

  • New: More accurate and faster Speech Recognition models.
  • Fixed: MXF files without duration information in the header can now be processed.
  • Improved: Simpler setup of local storage access for on-premise installations.

2025-12-16

API 1.63.0

  • Improved: Sessions are renewed automatically, so you stay logged in reliably.
  • Fixed: File type detection for files without an extension.
  • Fixed: "Token not provided" errors.

2025-11-10

Frontend 1.63.7

  • New: You can remove an uploaded file before you save a dictionary.
  • Improved: A security review of the web application.
  • Fixed: Changes to existing index collections are shown right away.
  • Fixed: Downloading artifacts.

2025-08-20

Manager 1.24.5 / 1.24.6

  • New: Visual Understanding returns structured answers that follow an output schema you define.
  • Changed: Speech Recognition no longer formats text into paragraphs by default.
  • Fixed: Model selection for Speaker Identification in the mining module wizard.

2025-07-31

API 1.62.0

  • New: A new login and account system with more secure handling of sessions and access tokens.
  • Fixed: You are logged out reliably.
  • Fixed: Missing thumbnails now return a proper "not found" response.

2025-07-21

Manager 1.24.4

  • New: Visual Understanding has new settings for shot detection and model selection.
  • New: Text Recognition has settings for language, threshold and mode.

2025-07-18

Frontend 1.63.0

  • New: Visual Understanding results are shown as formatted text or as structured data.
  • New: Mining module settings can accept structured (JSON) input.
  • New: Support for multi-tenant and on-premise deployments, with a new login flow.
  • Fixed: Long file names are shortened on upload, and you are told about it.
  • Fixed: After you submit a training, you are taken to the model overview.
  • Fixed: Images in the details of speaker index collections.

2025-06-18

Manager 1.24.1 / 1.24.3

  • Improved: Updated settings for Visual Understanding and Text Recognition.
  • Changed: Color Detection now returns up to 5 colours (3 by default).

Visual Understanding Worker 1.0.6

  • Improved: Text Recognition is now GPU-accelerated and processes faster.
  • Improved: Higher processing capacity for Visual Understanding and Text Recognition.

Looking for older entries? See the Changelog Archive for all changes up to June 2025.