LeVo

LeVo is a high-quality open-source AI song generation model from Tencent AI Lab that creates complete songs, including vocal performance and musical accompaniment, through multi-preference alignment. Built on the LeLM framework with an innovative dual-token modeling mechanism, it can process mixed tracks (vocals + accompaniment) and dual-track audio (separate vocals and accompaniment) in parallel, giving music creators more post-production flexibility.

Key Features:

• Full song generation: Automatically generates complete songs with vocals and accompaniment from lyric text, with 48kHz stereo high-fidelity output
• Zero-shot voice cloning: Copies a target voice's characteristics, including pitch, emotion, and rhythm, from just 3 seconds of reference audio
• Multi-dimensional music customization: Precise control across multiple dimensions including music style, gender, timbre, emotion, instrumentation, and tempo (BPM)
• Dual-track generation mode: Generates vocal and accompaniment tracks independently for professional mixing and post-editing
• Flexible deployment: Supports local deployment, Docker containerized deployment, and a Gradio web interface, requiring as little as 10GB of GPU memory

Use Cases: LeVo is widely used for short-video BGM production, song demo creation, game sound design, and podcast scoring. Music producers can use it to rapidly generate creative prototypes, content creators can easily add original music to their work, and AI researchers can build on its open-source code for deeper research into generative music.

Unique Advantages: LeVo rivals leading commercial models such as Suno 4.5 on multiple key metrics, significantly outperforming existing open-source academic models in musicality, audio quality, vocal-accompaniment harmony, and lyric alignment. As an open-source project from Tencent AI Lab, it provides complete code and pre-trained models free of charge to developers worldwide via GitHub and Hugging Face, advancing the adoption and innovation of AI music generation.