Local first
Core AI workflows process source media in the browser instead of sending it to an editing backend.
MIT licensed · No account required
AI video editing,
on your device.
A multi-track browser editor with local captions, multilingual voiceovers, vocal separation, background removal, and talking avatars.
Best experienced in a recent desktop version of Chrome or Edge.
Real timeline
Arrange visuals, captions, voiceovers, source audio, music, and overlays on time-accurate tracks.
Open stack
Built with ONNX Runtime, WebGPU, WebAssembly, WebCodecs, and the Web Audio API.
Inside the editor
One workspace, from source media to export.
Automatic captions
Whisper small q8 transcription with waveform-aware timing and editable, WYSIWYG subtitle styling.
Browser voiceovers
Multilingual Piper/VITS voices and Kokoro English generation, lazy-loaded when needed.
Source separation
Isolate vocals from the current source-audio clip and place the instrumental stem on the music track.
Talking avatars
JoyVASA audio-to-motion with LivePortrait neural rendering on compatible WebGPU hardware.
Vision-aware editing
Timestamped subject detection, smart crop, background removal, and caption avoidance for images and video.
Browser-side export
Preview the composed result and export MP4 or WebM without handing the project to a render server.
Current limitations
Experimental, and honest about it.
- Initial AI model downloads can be large.
- Browser memory and WebGPU support vary by device.
- 512 × 512 neural avatar rendering is still slow on current consumer GPUs.
- This is not yet a replacement for a mature desktop NLE.
Free, open source, and ready to try.