Eleven Music v2 is a generative audio model developed by ElevenLabs for the creation of high-fidelity music from natural language prompts. Released as a significant upgrade to the company's initial music generation technology, v2 focuses on enhancing vocal precision, instrumentation, and overall structural coherence. It is capable of generating complete compositions in 44.1kHz stereo, including synchronized vocals and instrumentation across a diverse range of musical genres.
The model is characterized by its ability to handle mid-track genre transitions, allowing a single audio file to shift between disparate styles—such as from orchestral to heavy metal—while maintaining a continuous musical flow. It also improves upon the rendering of complex vocal patterns, including rapid-fire rap and dense lyrical delivery. For structural control, the model utilizes a chunk-based composition plan, which allows for building a full song section-by-section (e.g., intro, verse, chorus, bridge) to ensure structural integrity over longer durations.
Key Features and Capabilities
Beyond standard text-to-music generation, Eleven Music v2 includes editing tools such as inpainting. This capability enables users to select specific segments of a track for regeneration without affecting the surrounding audio, allowing for fine-tuning of bridges or specific vocal lines. The model is multilingual, producing arrangements and lyrics that are designed to sound native across a growing list of supported languages, including English, Spanish, German, and Japanese.
The system is engineered to interpret both abstract descriptive language (e.g., "ethereal ambient soundscape") and technical musical terminology (e.g., "120 BPM synthwave with arpeggiated bassline"). For optimal results, users can specify structural elements and production quality descriptors in their prompts to guide the model's output more effectively.
Eleven Music v2 is trained exclusively on fully licensed data, ensuring that its outputs are cleared for commercial use in professional environments such as film, television, and advertising. The model aims to provide a consistent sonic identity, and it supports the creation of both vocal-led tracks and instrumental-only compositions.