Audio & video / Local FFmpeg tool
Save the soundtrack from a video as an MP3 at 96–320 kbps.
The first run loads an approximately 31 MB FFmpeg WebAssembly core from this site. It does not use a CDN or upload your files.
Files are limited to 200 MB total and 20 minutes; mobile devices use an 80 MB limit. Reverse, GIF and boomerang tools have lower duration caps.
At most 60 download files are generated.
Validated downloads will appear here after processing.
Method, example & boundaries
Extract the first audio stream from a local video and encode it as MP3 at a selected bitrate. Video frames are deliberately omitted, and completion is refused when the source has no audio stream, avoiding an empty or mislabeled download.
One approximately 12-minute MP3 containing audio only, with duration validated against the source.The video stream is not converted or retained; only the detected audio stream contributes to the MP3 result.
The local worker probes streams, requires at least one audio stream, maps the first audio stream, omits video, and runs the MP3 encoder at the chosen constant bitrate. It then probes the output before exposing the download.
Local AI / Speech to text
Open-source Whisper Tiny runs in your browser. The file, audio and result are never uploaded.
Choose an audio or video file in the workbench above first.
◎ The model loads only from 321Kit; MiniMax is not called and no transcript history is stored.
How it works
Only audio is written; the video frames are omitted from the result.
Only audio is written; the video frames are omitted from the result. Processing uses single-thread FFmpeg.wasm. The page does not upload, store or inspect file content outside this device. Cancel, error and page close terminate the worker and revoke temporary object URLs.
A source with no audio track cannot produce MP3 output. Files are limited to 200 MB total and 20 minutes; mobile devices use an 80 MB limit. Reverse, GIF and boomerang tools have lower duration caps.
Anonymous feedback
Found a problem or an idea to make this tool better? Tell us. We review every report and start working on actionable improvements within 2 days, continuously improving reliability and usability.