DrawSplatTM
Open Whiteboard
Released September 10, 2026
Record or import a video, generate English captions on your device, review the text and timing, then download SRT or burn the subtitles into your video.
Select a video or audio clip on the timeline and choose File → Generate subtitles…. After recording, you can also check Generate subtitles when adding this recording (English). Review the words and timing, then add captions to a new timeline track. Existing captions stay in place.
Choose Subtitles, load a video, and click Generate subtitles…. Edit the generated captions and download SRT, or choose Use subtitles for burn-in followed by Burn subtitles to MP4.
First use downloads a roughly 42 MB public speech model from Hugging Face and loads the app’s bundled speech engine. The download host sees normal connection metadata, but audio, video, and caption text are never uploaded. The model is cached when browser storage permits and can then work offline.
This release transcribes English speech, up to 30 minutes per clip and 512 MB per source file, using browser-decodable audio. Names, accents, noise, and music can affect accuracy. Silent files cannot be transcribed. Progress and cancellation controls are included.
VideoSplat now keeps Split, Delete, and other editing controls visible while you scroll. Timeline zoom −/+, plus Fit timeline, make long sequences easier to navigate without changing the video canvas or output resolution.
The recorder remembers your microphone choice, shows an audio meter, and lets you play a recording with sound before adding it. Export reports audio connection failures instead of silently saving video without the expected audio.
The generation dialog downloads SRT with times relative to the selected clip. After adding captions, File → Save captions as SRT exports project-relative times. Captions are editable timeline clips; moving a source video later does not automatically move its caption track. Keep Burn in subtitles enabled when exporting to make captions permanent.
All twelve self-host ZIPs are refreshed. MediaSplat and VideoSplat packages include the local speech engine and shared source package; the speech model downloads on first generation. Keep the included folders and runtime assets together.
Verified with real speech transcription in both apps, cached offline transcription, editing and SRT downloads, silent-audio rejection, cancellation, recording-triggered generation, and a MediaSplat MP4 containing edited generated captions.
Release notes · GitHub release and checksums · Earlier subtitle import and burn-in walkthrough