⚡Ultra-Fast GenerationUnder 2s per full song on A100, under 10s on RTX 3090. Supports 10s to 10min audio generation.
🎵Commercial-Grade Quality1000+ instruments and styles, 50+ languages with lyrics prompt for structure and style control.
🎛Versatile EditingCover generation, repainting, track separation, vocal-to-BGM, multi-track generation, and more.
🚀Lightweight & CustomizableRuns locally with <4GB VRAM. Train a LoRA from just a few songs in 1 hour on a 3090.
🎤Reference Audio InputUse reference audio to guide generation style. Create covers from existing audio.
🔬Audio UnderstandingExtract BPM, key/scale, time signature and caption from audio. Auto-generate LRC timestamps.