AI Song Generator

Yes — the render is a single piece of audio with sung vocals over an instrumental, not a MIDI melody you have to voice yourself. It is a generative model, though, so phrasing and pronunciation vary from take to take, and an awkward line is usually fixed by re-rolling with a different seed rather than by rewriting. Leave the lyrics box empty, or tick Instrumental, to get a vocal-free track.

From 10 seconds up to 5 minutes on this server, with 2 minutes as the default. A full song or a cover costs 5 credits, extending a track 4, and repainting a single segment 3 — the live price is shown on the generate button, since an operator can retune it. If the render fails, your credits are refunded automatically.

Yes, and it's tighter than the character count suggests: the model has a lyric window of roughly 2,000 tokens, which dense scripts reach well before any character limit. Longer lyrics are trimmed at a line boundary before the render rather than being cut mid-word. Nothing is deduplicated, reordered or filtered — every line that fits is passed to the model exactly as you typed it, repeated choruses included.

Yes. Cover restyles a reference track you upload, Repaint regenerates a chosen segment of it, and Extend continues a partial track. All three take an audio upload of up to 30 MB. There's also a 'surprise me' mode where you give a one-line brief and the model writes the style description and lyrics for you.

The audio in your chosen format — MP3, WAV, FLAC, Opus or AAC. For songs with vocals you also get synced lyrics generated from the finished audio, downloadable as LRC, SRT or VTT, and a one-click hand-off that sends the track and its timings straight into the Karaoke Video Creator.