YuE2-3B

An open music generation model from Multimodal Art Projection that produces full songs with vocals and accompaniment from text and lyrics, delivers song quality competitive with Suno v5/v6, and sets a new SOTA on WildSongBench. It supports zero-shot cover generation and symbolic melody/chord editing, including an agentic editing workflow, while running locally on a single 24 GB GPU at 48 kHz stereo.