4 May 2026

Sharper Profanity Handling and Faster Transcription

Improved profanity audio mode handling in exports and playback.
The ‘beep’ sound effect is now generated at the exact length needed to keep file sizes small.
The ‘beep’ is also cached locally so it can be reused across projects rather than generating a new one each time.

When instances of profanity are close together, we no longer pre-apply a crossfade, which can distort the audio of adjacent words.

Improved playback readiness detection so videos now load faster when switching between projects or catalogs.

Transcription behavior now respects project length to optimize model loading across ANE, CPU, and GPU.
Deleting transcriptions is now much faster.
Importing transcripts is now faster thanks to a new baseline architecture in this release.
We now use a beam cursor to indicate the current word, and it is now more precise.
The model cache is now used consistently. Previously, it was sometimes ignored.

General cleanup of edge cases and improvements to internal consistency.
Fixed cursor behavior inconsistencies.
Improved shutdown experience with progress indication.
Scrolling in the transcription pane is now smoother when many filters are applied.
Tooltip overlays now align more precisely with their targets and are styled more consistently throughout.
Database speed has been improved on first load.
We no longer display raw errors when an export fails.
Sequence no longer holds the model in memory, so RAM usage remains low until transcription is triggered, after which the model is unloaded to free up system memory.
Cached audio files are now cleared when Sequence closes to protect your data privacy.

Find the Story. Build the Rough Cut. Keep Creating.