⚠ When lyrics are provided, auto-generation is disabled. Use structure tags: [verse] [chorus] [bridge]
⚠ Music now runs on YuE2/ComfyUI (GPU-queued) and takes noticeably longer per song than before — capped at 4 per playlist.
Click to choose an audio file (mp3/wav/flac)…
Extraction listens to the reference's vocals and rewrites them to fit its length. Instrumentals fall back to new lyrics.
Pick it if the reference is sung — auto-detect often gets sung Farsi wrong.
⚠ Melody is transcribed from the reference audio via SheetSage2; lyrics/vocals are re-sung by YuE2.
The same lyrics are sung with different random seeds — some come out as clear singing and some as gibberish, so you pick the best one. Each version takes about a minute.
⚙ Settings
Saved settings persist on the server (survive redeploys) and take effect immediately for the next generation — no restart needed.