ClipCatalog logoClipCatalog
EN

Changelog

Every update, from the latest release to the very first beta.

v0.22.1

2026-10-02

Highlights

  • Error reports no longer flood your connection. A damaged video could make FFmpeg produce megabytes of error output, which ClipCatalog then tried to upload every minute without success, losing that session's error reports. Reports are now trimmed to a sensible size and sent in smaller batches, so they get through.
  • Better privacy when a video cannot be read: GPS positions, device serial numbers and other details stored inside your videos no longer end up in ClipCatalog's log files, in diagnostics exports or in error reports. Values from your catalog are also kept out of database error messages.

Changes

  • •Fix video decoding and AI processing using a different graphics adapter than the one selected, on some PCs with more than one adapter
  • •The download screen now explains that the larger free-space figure is only needed while downloading, and lists the storage folders under "Where the files go"
  • •The installer now uses European Portuguese throughout, matching the rest of its Portuguese text

Your library keeps getting better

A short note when we ship something worth grabbing — new features, faster processing, and fixes. About once a month, never more. Just product updates, no spam.

v0.22.0

2026-09-29

Highlights

  • The one-time face regrouping is now much safer to interrupt. Stopping it, or a crash partway through, no longer throws away the people it has already grouped: the next run continues where it left off. Stop now cancels the work in progress right away, and very large libraries get enough time for each grouping step instead of timing out.
  • When your graphics card stops responding during AI processing, ClipCatalog now pauses GPU acceleration on that card for the rest of the session and carries on with the CPU, instead of trying the card again at every start. A notice you can dismiss explains what happened, and Settings offers Try GPU again. PCs with integrated graphics and limited RAM also get a heads-up during automatic folder scans.
  • Export diagnostics from Settings: save ClipCatalog's log files and details about this PC to a zip file you can send to support. Log lines now carry the date, and more log history is kept.

Changes

  • •Videos that keep failing face detection are set aside after three attempts and listed with the folder's skipped videos, where you can give them another try
  • •Fix the source frame shown for faces in 360° videos, and keep faces on their own frame when a video's thumbnails are rebuilt
  • •Keep face images when rebuilding a video's thumbnails fails or is stopped
  • •Fix a thumbnail sometimes failing to generate near the end of a video, or when a slow seek had to be retried
  • •The backup card now says why a backup was postponed: a running sync, background indexing, or low disk space. If the catalog check at startup times out, Check again runs a longer check
  • •The upgrade welcome shows how many videos each step will re-process, and the startup update screen now says that closing the app restarts the update from the beginning
  • •Relocate and Subfolders to sync now handle folders nested deeper than Windows' 260-character path limit
  • •Fix parts of some file names on network shares appearing unredacted in error reports
  • •Error reports keep the original cause of a transcription startup failure and say which processing stage was interrupted
  • •Remove an unused 3 MB geolocation database from the bundled ExifTool

v0.21.1

2026-09-15

Highlights

  • A slow face check no longer stops the whole sync. On PCs where the AI runs on the processor instead of a graphics card, verifying the faces in a single video could take longer than ClipCatalog allowed and abort the entire sync, so image search, tagging and transcription never ran either. Face verification now tolerates the occasional slow video, and if it does give up it ends only the face step — everything else keeps going, and the affected videos are retried on the next sync.
  • Folder scans now tell you which subfolders could not be read. Watching a whole drive could leave a scan permanently incomplete because of a single protected system folder, silently re-scanning the same folder over and over. ClipCatalog now names the subfolders it could not read and suggests unchecking them under "Subfolders to sync", and it no longer counts protected or already-excluded folders as errors.
  • The download and storage screens now show where your data goes. Before downloading the AI models, you can see each storage location on your PC, what is kept there, how much free space it needs and how much is available, with a copy button for every path. Low-disk warnings name the drive that is actually full and make clear that your original videos stay where they are.

Changes

  • •Fix video processing parking indefinitely when starting FFmpeg, FFprobe or ExifTool stalled — a sleeping disk, a disconnected network share, or an antivirus scan of the program files
  • •Fix automatic scans losing a full scheduled interval when a drive disconnected mid-scan; a missed drive is no longer counted as an attempt
  • •Add stall detection to thumbnail extraction, so a stuck extraction is recorded in the logs instead of sitting silently for hours
  • •Name the two free tiers apart throughout the app: the never-expiring free trial covering your first 500 videos, and the 14 days of full library access
  • •Fix incorrect advice telling free-trial users their catalog key is backed up — it is not, and the catalog cannot be recovered unless you export the key yourself
  • •Add precise error messages for activation and recovery failures, including protected device key and protected storage problems, and indicate which ones are worth retrying
  • •Fix identity recovery being unreachable when you no longer have your recovery link
  • •Better handling of a full disk during sync: a catalog storage shortage is now told apart from a genuine disk error
  • •Add a distinct warning when face verification times out for a single video, instead of wrongly reporting that face detection failed
  • •Improve the guidance shown when AI model verification fails, including when the affected files cannot be removed automatically
  • •Fix the image embedding service continuing to process a batch nothing is waiting for anymore
  • •Update whisper.cpp to v1.9.4

v0.21.0

2026-09-12

Highlights

  • Activation works again on PCs whose security chip is present but not working. If the TPM is unprovisioned, locked out, or waiting on a firmware update, ClipCatalog now protects its device key in software instead of failing outright, so your license activates and the "Protected device key unavailable" banner no longer appears on every launch. PCs with a working TPM keep using it.
  • Face analysis now explains itself. The People screen tells you exactly where it stands (not analyzed yet, analyzing, no usable faces found, or faces found but not yet grouped), and a new Analyze now action starts the work on the folders that are still waiting, from the People screen or straight from a directory card.
  • Filter by folder straight from a search result: click any folder name in a result's path to narrow the results to that folder, and click it again to remove the filter.

Changes

  • •Directory cards now show when face detection is off, with a one-click Turn on, and offer Analyze now for folders whose videos haven't been checked for faces yet
  • •Analyze now is honest about what it can't do: it says when the videos are on storage that isn't connected, when a sync is already running, and when a start request didn't take effect
  • •The video path filter now explains that it matches anywhere in the folder path and points file-name searches at their own filter, with a dismissible hint and clearer examples
  • •Record the Windows status behind a failed device-key operation in the log and in error reports, so a security chip that is present but faulted can be told apart from one that is simply missing

v0.20.1

2026-09-10

Changes

  • •Fix the license screen reporting a busy licensing service, or asking you to email support, when this computer's clock is set behind the real date and time. It now says so and offers Check now; your license, catalog and videos are untouched
  • •Record in the log why a license credential was rejected, so a licensing problem can be diagnosed without error reporting turned on

v0.20.0

2026-09-09

Highlights

  • Face recognition rebuilt end to end — faces are now detected from 720p crops instead of low-resolution thumbnails, through a GPU-accelerated streaming pipeline, with much better grouping so one person is far less likely to be split across several entries.
  • A dedicated People screen — name and curate everyone in your library, review merge suggestions as whole-identity group cards, and undo any merge from a full merge history.
  • Your own tags — add and edit tags on any video, in bulk if you like. The app remembers tags you've used and suggests them as you type, and they're searchable alongside the AI-detected ones.
  • Tag confidence and a strictness filter — the visual tagging model now reports a real confidence for every tag. Dial strictness up for precision or down for recall, and see which tags actually matched right on the thumbnails.
  • Copy to folder — select any set of search results and export copies to a folder of your choice, with live progress and cancel.
  • Scheduled folder scans — give each directory its own scan schedule so new footage gets indexed without you remembering to run a sync.
  • Transcription now runs on older CPUs — the bundled whisper build no longer requires AVX2, so machines from before roughly 2013 (Intel) or 2015 (AMD) no longer crash instantly at transcription. A new stall watchdog also recovers a wedged transcription instead of hanging the whole sync.
  • Your catalog is protected — export your database key, back up on demand, an automatic safety copy before every database upgrade, and recovery when the database is damaged.

Changes

  • •Search filters reworked — folder-only path matching, filename wildcards, a Match case toggle, per-field tooltips, and unified match highlighting
  • •Matched tag chips now sort first and highlight in match green, with a "+N" overflow counter and a corner play pill on thumbnails
  • •New floating selection action bar and a slimmer sort header that measures its own breakpoints
  • •Expand a results section to larger thumbnails, with a "+N" veil, a Show-fewer tile, and a sticky collapse strip
  • •Fix tag chips that rounded to "0%" — they now fall back to an occurrence count
  • •People screen additions — extract faces into a new or existing person, delete a person in one click, rename by double-click, run "Find more matches" on demand, and reveal the source video a face came from
  • •False detections (formerly "junk faces") get their own review and trash view with batch rescue, and the pipeline now suppresses non-face clusters on its own
  • •Two people in the same frame are never merged or suggested as the same person
  • •Per-directory "Re-sync folder for HD faces" nudge, so older folders can be upgraded to 720p face detection at your own pace
  • •Sync survives sleep — a wake-lock holds the machine awake during a run, and transcription resumes cleanly after a suspend
  • •Low disk space is detected up front — sync stops with a clear warning instead of filling the drive
  • •Skipped videos are now a proper terminal state, with a dialog organized by reason and a direct remedy on every card
  • •Failed sync runs, stage-fatal errors, and sidecar startup failures now surface honestly instead of looking like a silent stop
  • •A directory that fails three scheduled scans in a row is held rather than retried forever
  • •Transcription mass-failure breaker — a run that starts failing every video stops instead of burning through the library
  • •Faster metadata stage via an adaptive concurrency ramp and batched database writes
  • •Much faster semantic-index ingestion, plus a fix for a race that left freshly-indexed videos silently unsearchable
  • •GPU keyframe-batch thumbnail extraction, keyframe-snap sampling for long videos, and a dedicated CPU lane for long 10-bit and HDR footage
  • •360 thumbnails — fixes for short clips, zero-thumbnail failures, tail corruption, and dual-fisheye handling
  • •Fix thumbnails losing their rotation on hardware-accelerated paths
  • •Fix thumbnail generation against FFmpeg 8.x
  • •Thumbnails are never wiped on abort or failure, and database-to-disk drift now self-heals
  • •Semantic search reworked around native result grouping and tiered candidates, with FP32 vision embeddings for better matches
  • •Fix semantic search getting stuck when started while the AI models were still downloading
  • •One-time offline Upgrade run with an app-global progress banner you can pause, resume, or stop
  • •Automatic safety copy before every database upgrade, plus a progress screen instead of a frozen splash
  • •30-day offline access — keep working for up to 30 days without an internet connection
  • •Activate a license inline during the trial, move your license to another PC yourself, and deactivate this device from the license screen
  • •Fix paid licenses being downgraded to trial by a transient storage fault, a system clock running behind, or a stalled startup check
  • •In-app security advisories — get notified when your installed version is affected by a published vulnerability
  • •Published CycloneDX software bill of materials, an end-of-support notice, and an auto-update opt-out
  • •EU AI Act transparency disclosures for face recognition and AI features
  • •Silent and unattended installs — the installer's EULA and Terms clickwrap moved to the app's first-run agreement gate, which also enables winget distribution
  • •New logo and brand mark across the app, installer, and website
  • •Updated bundled components — whisper.cpp 1.9.2, FFmpeg (LGPL) N-125444, and the database stack on drift 2.34 with SQLCipher 4.17
  • •Every snackbar replaced with a dialog or an in-place confirmation, so nothing important disappears on a timer
  • •Directory health folded into the header status box, with the last-sync date carrying staleness
  • •Settings storage totals are instant (they used to take 20+ seconds), and cross-screen links now scroll straight to the right section
  • •Localization pass — European Portuguese consolidation, German formal address (Sie) throughout, Russian, French and Portuguese plural fixes, Arabic numeral handling, and native-speaker review across the licensing, face, and upgrade string families
  • •Various polish: quieter logs, a fix for the "Channel was closed" error flood at shutdown, fewer crashes from malformed child-process output, and numerous smaller UI and copy fixes
  • •Upgrading from an earlier version now shows the exact download size up front (about 2.8 GB of updated AI models and tools) and checks free disk space before starting

v0.19.1

2026-05-12

Highlights

  • New first-run consent screen for anonymous error reporting and telemetry. Easy to change anytime from Settings, with clearer wording about what's collected and why.
  • Faster CPU tagging when the GPU isn't available. The visual tagging sidecar now applies all ONNX Runtime optimizations on the CPU path, so automatic GPU-to-CPU fallback is noticeably quicker.
  • Redesigned third-party licenses screen with a categorized layout and per-component license text, plus offline copies of every LICENSE file under the install directory for audit-friendly access.

Changes

  • •Smoother 14-day trial start and exit experience
  • •Email templates and the license screen now clearly state the 500-video search limit that applies after the trial ends
  • •Native-speaker translation polish for the credits and licenses screens (French, Spanish, Portuguese, Russian, Korean)
  • •Portuguese translations consolidated to European Portuguese (pt-PT) for consistency
  • •Bumped Python sidecar dependencies across tagging, face, text, and vision sidecars
  • •Improved tagging-sidecar error reporting
  • •Disabled processing stages on the directory sync view now show a clear 'Off' label with a tooltip explaining where to enable them, replacing the previous misleading 0%

v0.19.0

2026-04-30

Highlights

  • Free 14-day trial extension — enter your email on the license screen for a one-time extension. The license screen has also been redesigned to show what's been indexed and what's still waiting.
  • Faster, higher-quality transcription with the new whisper q8 model — about 5x faster on CPU and noticeably better on noisy audio. The ~850MB model will download automatically on first run; the previous model is removed to free disk space.
  • Per-stage GPU overrides — pick a different GPU for tagging, face detection, transcription, and vision. Helpful on systems with both an integrated and a discrete GPU.

Changes

  • •Significantly faster thumbnail generation by running multiple full-decode paths in parallel for 4K and lower videos
  • •Better sync progress feedback — realtime processing speed, accurate ETAs, percentage progress per stage, and smoother updates
  • •Default to integrated GPU for transcription where supported — much faster on recent integrated GPUs
  • •Add in-app review prompt with optional anonymous feedback
  • •Rework benchmark results dialog with per-workload pills and an apply-preview block
  • •Improve CPU fallback — vision sidecar now correctly falls back, hybrid CPU/GPU mode is shown distinctly from pure CPU fallback, and the warning wording is clearer
  • •More resilient transcription — recover when whisper-server crashes mid-job (no more lost videos), probe the GPU with exponential backoff to detect when it comes back, and stopping a sync during transcription now cancels in-flight FFmpeg WAV extraction
  • •Fix 360 thumbnail generation failing for 10-bit HEVC videos (DJI Osmo 360, Insta360 HEVC Main 10)
  • •Fix FFmpeg hardware acceleration not being used after upgrading from versions where GPU acceleration was disabled in the old settings
  • •Fix paid licenses being silently downgraded to trial when the license token expired while offline
  • •Smoother in-app updates — the installer now waits for the running app to close instead of racing the file copy
  • •Various polish: clearer processing-event wording, better tail-corruption detection during the thumbnail stage, restored hand cursor on the top navigation bar, improved check-for-updates reliability, and translation fixes for the benchmark button and preview-thumbnails label

v0.18.1

2026-04-16

Highlights

  • Dropping Windows shortcuts (.lnk files) and Network Locations into the app now correctly resolves the target path, so you can add directories via shortcuts from your Desktop or Explorer's "This PC" network locations.

Changes

  • •Fix adding a folder that was previously added but has since been renamed
  • •Fix directory re-link when the directory contains duplicate video files
  • •Fix directory re-link falsely reporting hidden files (e.g. GoPro chapter files) as new videos
  • •Fix spurious processing warning when stopping sync during the vision stage
  • •Add SQLCipher license notice to the credits screen

v0.18.0

2026-04-09

Highlights

  • NAS and network drive support — you can now add directories on NAS devices and mapped network drives (Windows UNC paths and macOS network volumes)

Changes

  • •Fix NAS volume not updating to unavailable status on disconnect
  • •Fix directory merge issues on mapped network drives (path separator normalization)
  • •Fix showing "whole volume" instead of the selected subfolder in the relink popup
  • •Fix total video size and duration not updating on the directory screen after metadata analysis
  • •Fix multiple localization issues in add directory and volume popups
  • •Reduce log noise from corrupted video files (downgraded from error to warning)
  • •Fix NAS volume polling causing log spam and UI hangs on disconnect

v0.17.9

2026-04-05

Highlights

  • Significantly improved 360 video thumbnail generation performance (10-30% faster for DJI and Insta360 cameras)
  • Volume type (internal, external, remote) is now shown in the directories screen and volume info overlay. This requires running the "Synchronization" operation so the app can determine the volume type for existing volumes.

Changes

  • •Fix 360 thumbnail scheduling and full-decode failures
  • •Fix issue where low-res thumbnails could not be created during thumbnail generation
  • •Stopping a sync is now more responsive during the thumbnail stage
  • •Improve localization on the license screen (add missing translations)
  • •Better error handling for files that disappear during sync

v0.17.8

2026-03-29

Highlights

  • Significantly improved search results scrolling performance and responsiveness

v0.17.7

2026-03-27

Changes

  • •Fix hardware device (GPU) selection not persisting correctly between sessions
  • •Fix thumbnail processing issues
  • •Improve memory usage by scoping exif pool to metadata stage only

v0.17.6

2026-03-26

Changes

  • •Improve UX when launching versions 0.17.1 and later when the AI models had already been downloaded. In this case, they should not need to be downloaded again.

v0.17.5

2026-03-26

Changes

  • •Fix thumbnail paths mismatch introduced in the last release

v0.17.4

2026-03-26

Highlights

  • The app can now be installed for multiple users on one computer. Every user gets their own database and trial mode individually.

v0.17.3

2026-03-24

Highlights

  • This version contains small bug fixes

v0.17.2

2026-03-21

Highlights

  • Processing errors are now replaced with a more detailed Processing Events log with severity levels, giving better visibility into what happened during synchronization

Changes

  • •Fix Natural Language Search thumbnail borders to correctly indicate matches
  • •Improved GPU device settings and benchmark accuracy — CPU fallback no longer inflates GPU performance numbers
  • •Improved stability when stopping synchronization during metadata analysis
  • •Various sidecar stability and error handling improvements

v0.17.1

2026-03-18

Highlights

  • Natural Language Search result thumbnails get better highlighting

Changes

  • •improvements around app stability and UX
  • •Hardware acceleration benchmark can now properly be skipped

v0.17.0

2026-03-15

Highlights

  • New feature: Natural Language Search Filter. Describe what you are looking for in natural language and configure how strict you want to be about this. Requires re-processing of videos to make them discoverable in the search
  • More streamlined hardware acceleration settings with more control over each processing stage and a benchmark option which finds the best performance for visual content detection and transcription
  • Performance improvements for thumbnail generation and metadata analysis stages

Changes

  • •use webp for thumbnails in the future, roughly 20% less space consumption without quality loss
  • •highlight score computation improvements to work better with big libraries
  • •small bugfixes and UX improvements

v0.16.0

2026-03-01

Highlights

  • Application shutdown UX improved
  • The Transcription benchmark dialog now allows the user to skip a device he doesn't want to have considered anyway
  • Fix thumbnail artifacts when processing 8K videos on a weak GPU

Changes

  • •The installer now properly uses the system language
  • •progress indicators during sync now always show total percentage, not current work percentage which is much less confusing
  • •various small UX improvements

v0.15.0

2026-02-28

Highlights

  • Improve app startup time by up to 6 seconds
  • Add Splash Screen on app startup

Changes

  • •add tooltip for synchronization stop button to indicate that no progress is lost
  • •fix GPUs not getting detected in some cases
  • •the search results now offer the expand button and the "find more videos with faces from this video" button even when the volume or directory is not available

v0.14.1

2026-02-24

Changes

  • •fixed the automatic folder re-linking / relocating when the user has moved or renamed the path to a directory which he added as a video directory into ClipCatalog

v0.14.0

2026-02-22

Highlights

  • Improved the performance of search against big video libraries greatly when common detected content labels or words were used
  • Removed the limitation of 32000 videos in the video library - now you can have an even bigger library

v0.13.6

2026-02-22

Highlights

  • the changelog is now machine translated into all supported languages
  • the changelog now also shows on the website

Changes

  • •Improvement: Search by spoken words filter: when entering multiple words separated by a space then the app will split the input into multiple words at the spaces

v0.13.5

2026-02-21

Changes

  • •installer is now available in 10 different languages
  • •installer auto-picks the current language, shows language selector otherwise

v0.13.4

2026-02-21

Changes

  • •Improve the video duration filter so lower durations (<60seconds) can be selected much easier
  • •Add a hint to the spoken words filter to indicate that the user needs to enter a single word
  • •Auto-split words at the spaces in the spoken words filter

v0.13.3

2026-02-21

Changes

  • •improve the UI when downloading assets at first app start

v0.13.2

2026-02-20

Changes

  • •minor bugfix in initial asset download screen

v0.13.1

2026-02-20

Changes

  • •small bugfixes

v0.13.0

2026-02-20

Changes

  • •Bugfix: In-App Update functionality
  • •Bugfix: In-App Video Player causing video stretching

v0.12.1

2026-02-19

Changes

  • •fix links to our privacy document

v0.12.0

2026-02-10

Changes

  • •First public release

v0.11.1

2026-02-10

Changes

  • •Improve the updater
  • •update indicator
  • •auto check for new version on app start

v0.11.0

2026-02-09

Changes

  • •Re-link functionality for moved directories
  • •Bugfixes with search filters and non-existent videos

v0.10.3

2026-02-08

Changes

  • •Fixed installer packaging to include newly added binaries
  • •Improved application startup error handling

v0.10.2

2026-02-06

Changes

  • •Upgraded Flutter to a newer version
  • •Made highlight scores available only after videos finished processing
  • •Made footage type available only after full processing and with face detection enabled

v0.10.1

2026-02-06

Changes

  • •Improved footage type calculation
  • •Improved highlight score calculation

v0.10.0

2026-02-05

Changes

  • •Added selection of the transcription backend (CPU can be faster than GPU in some cases, especially on iGPUs)
  • •Improved the visual design of the Directories screen
  • •Improved performance of the Directories screen

v0.9.8

2026-02-04

Changes

  • •Improved overall performance

v0.9.7

2026-01-30

Changes

  • •Released an internal testing build

v0.9.6

2026-01-30

Changes

  • •Fixed minor bugs

v0.9.5

2026-01-30

Changes

  • •Improved the directory filter user interface

v0.9.3

2026-01-30

Changes

  • •Improved the update process
  • •Improved asset downloading to fetch only missing or outdated files

v0.9.2

2026-01-30

Changes

  • •Optimized metadata processing

v0.9.1

2026-01-30

Changes

  • •Added a GPU selection option
  • •Improved thumbnail extraction performance

v0.9.0

2026-01-29

Changes

  • •Released a beta version