Help Centre

Open the app, describe what you want in a sentence, and generate. A useful first description names a genre, a tempo and whether you want vocals — "slow lo-fi beat, 75 BPM, dusty piano, no vocals" will get you further than "chill music". Generation takes under a minute; regenerate as many times as you like and keep the take you want.

Five things, in rough order of impact: genre, tempo, mood, instrumentation, and whether there are vocals. Tempo changes the feel more than anything else, so name a BPM if you have one in mind. Naming instruments you do not want ("no drums") works as well as naming ones you do.

Standard generations run up to four minutes. Longer pieces are built by generating sections and extending, rather than asking for one continuous take — that keeps the structure coherent instead of drifting.

Yes. Paste them in and they will be performed rather than rewritten. If you want the lyrics generated first so you can edit them before committing to audio, use the lyrics generator and hand the result over when the words are right.

Ask for one explicitly — "instrumental, no vocals" — or switch the vocal toggle off before generating. Generating an instrumental from the start gives a cleaner result than generating a song and removing the vocal afterwards, because nothing has to be cancelled out.

Generation is probabilistic — the same description explores a different path each time. That is useful: run the same prompt several times and pick the best take. If you want consistency across a set, keep the prompt identical and change only one variable at a time.

Generated tracks download as WAV for full quality and MP3 for sharing. The free browser tools on this site always export WAV, because encoding MP3 in-browser would mean a second lossy pass on audio that has already been compressed once.

No. Every tool on this site — vocal remover, stem splitter, trimmer, joiner, BPM and key detection, converter, compressor — runs entirely in your browser using the Web Audio API. Your file is decoded locally and never sent anywhere. You can confirm this by opening the network tab while you use one.

The browser tools cap uploads at 20 MB. Because processing happens in memory on your own machine, very large files can exhaust the tab rather than fail cleanly. Trim the section you need first, then run the tool on that.

Yes. Tracks you generate are yours to use commercially, including in paid advertising, with no per-play royalty and no attribution requirement. That covers video, podcasts, games, streams and client work.

Your own generations are not registered with Content ID by us, so they should not be claimed. If you do receive a claim, it is usually because another party registered similar or identical audio; dispute it and state that the track was generated and licensed to you.

You need a mechanical licence for the composition, exactly as any covering artist does — generating the recording does not change that. The recording you create is new and yours; the underlying song is not.

You hold the usage rights to what you generate under your plan. Note that copyright registration for purely machine-generated work is unsettled and differs by country — in the US, the Copyright Office requires human authorship for registration. Editing and arranging the output strengthens any claim you make to it.

Plans are managed in account settings inside the app. Changes take effect at the start of the next billing period, and cancelling leaves everything you have already generated available to download.

Request deletion from account settings. Download anything you want to keep first — deletion removes your generations along with the account, and it cannot be reversed.

Browsers block audio until you interact with the page, so press play rather than expecting sound on load. If it is still silent, check the tab is not muted and that the correct output device is selected. On the separation tools, a silent result usually means the file was mono — the tool warns about this before processing.

Reverb and delay on a voice are spread across the stereo field rather than centred, so they survive even when the dry vocal is cancelled. Doubled or hard-panned backing vocals survive too. Narrow or heavily limited masters give the technique little to work with — the stereo width reading on the tool tells you what to expect before you process.

The tools accept anything your browser can decode — MP3, WAV, M4A/AAC, OGG, FLAC. Files that fail are usually renamed rather than converted (an M4A saved as .mp3), corrupt, or DRM-protected. Re-exporting from your player or editor generally fixes it.

Everything runs on your own machine, so a long file on a modest laptop takes real time. Trim to the section you need first. If a tab stops responding entirely, the file is likely close to the 20 MB cap — reload and try a shorter excerpt.

Retry first — most failures are transient. If it repeats, shorten the prompt; extremely long or self-contradictory descriptions ("fast slow ambient death metal lullaby") can fail to resolve. If failures continue across different prompts, it is on our side rather than yours.