YuE2

What YuE2 Does

YuE2 is an open-weight AI music generation model that transforms written lyrics and style descriptions into complete songs with vocals and instrumental accompaniment. Unlike traditional music generators that produce audio directly, YuE2 uses a unique two-stage approach: it first composes an editable musical score in ABC notation, then renders that score into audio. This symbolic planning workflow separates composition from performance, giving users the ability to inspect and modify melody and chord progressions before final synthesis.

Who It's For

YuE2 serves three distinct audiences:

• Songwriters who have lyrics but need a vocal demo to hear their words performed with accompaniment
• Music producers who want to work directly on the composition itself—inspecting, revising, and iterating on melody and harmony at the score level
• Developers and researchers who need an open music model with downloadable weights, published evaluation materials, and separate planning and synthesis interfaces for custom workflows

Key Features & Capabilities

• Symbolic Planning Workflow: Generates an editable ABC notation score before audio synthesis, making melody and chord symbols explicit and modifiable
• Three Planning Modes: Supports melody-and-chord planning, melody-only planning, and generation without a symbolic plan (via official API)
• Reference Audio Generation: Use uploaded recordings or browser-recorded audio (up to 120 seconds) to guide new renditions in different musical styles
• Multi-Language Support: Generates songs in English, Chinese, Japanese, Spanish, and other languages
• Audio Transcription: Convert existing recordings into downloadable ABC notation files and score images
• Multiple Export Formats: Download generated music as MP3, WAV, or FLAC
• Agent-Assisted Editing: Official workflow demonstrations show how external agents can help revise scores, styles, and lyrics through multiple musical iterations

How It Works

Users paste lyrics into the browser-based studio, describe the desired musical style (genre, vocal character, instruments, mood), and optionally upload reference audio. The model generates a complete song in approximately 60 seconds. Text-to-music generation uses 3 credits; music-to-music (with reference audio) uses 4 credits. In the official model workflow (outside this studio), users can access the intermediate ABC score, modify chord progressions or melodic phrases, and regenerate audio from the edited composition.

Benchmark Performance

In the YuE2 team's published September 2026 comparison against Suno v6 across 192 WildSongBench prompts, YuE2 scored higher on musicality (5.9075 vs 5.6558) and SongBench average (6.7316 vs 6.5562), though Suno v6 showed fewer pronunciation errors. This positions YuE2 as a strong choice when musical composition quality is the priority, with the caveat that pronunciation accuracy requires careful review.

What Makes YuE2 Different

The defining distinction is access to the composition layer. While most AI music generators deliver only finished audio, YuE2's symbolic planning stage makes the musical score an explicit, inspectable, and editable artifact. This approach appeals to users who care about what the song is made of—the actual notes, chords, and structure—not just how the first take sounds. The model's open-weight architecture and published evaluation materials also make it a valuable resource for researchers investigating AI music generation.