Generate music

What you can do

Generate full songs with vocals and lyrics from a text prompt.

Create instrumentals across any genre — electronic, orchestral, jazz, lo-fi, rock, and more.

Use a reference track to guide the style of your generated music.

Generate sound effects and ambient audio alongside music.

Models we recommend

For full songs with vocals

MiniMax Music 1.5 is the strongest option for generating complete songs with vocals. It produces tracks up to 4 minutes long with natural-sounding singing in English and Chinese. You can write lyrics with structure tags like [verse], [chorus], and [bridge], and optionally upload a reference track (5–30 seconds) to guide the style. Great for songwriting, demos, and soundtrack prototyping.

ElevenLabs Music generates studio-grade songs up to 5 minutes from a text description. Toggle between vocal and instrumental output with force_instrumental. It handles detailed composition plans — specify genre, mood, tempo, instrumentation, and song structure in your prompt. A good pick when you want polished, ready-to-use tracks without writing lyrics yourself.

For instrumentals and background music

Stable Audio 2.5 generates high-quality instrumentals and sound effects up to about 3 minutes from text prompts. It also supports audio inpainting and continuation — feed it a clip and it'll extend or fill gaps seamlessly. Open-source weights mean you can self-host it if you need to.

ElevenLabs Music with force_instrumental set to true produces clean instrumentals across any genre. Its natural-language prompting makes it easy to describe exactly what you want — "relaxing ambient pads and piano for meditation" or "high-energy synthwave with retro 80s vibes."

Google Lyria 2 produces 30-second clips at 48kHz stereo — the highest audio fidelity in this collection. It supports negative prompts to exclude unwanted elements. Best for short, high-quality loops, jingles, and audio samples where pristine sound quality matters most.

For style-guided generation

MiniMax Music 1.5 lets you upload a reference audio clip (5–30 seconds) and control how much it influences the output with a style strength parameter (0.0 to 1.0). This makes it ideal for creating variations on an existing track or producing music in a specific artist's style.

MiniMax Music 01 is the predecessor — faster and simpler, generating up to 60 seconds from a reference track and lyrics. Good for quick experiments and shorter clips.

For open-source and self-hosting

Stable Audio 2.5 is built on Stability AI's open-source stable-audio-tools and weights are publicly available. Run it on Replicate or deploy it on your own hardware.

ACE-Step is an open-source model that generates full songs with vocals in about 20 seconds on an A100. It uses a diffusion-based architecture that's 15× faster than autoregressive approaches. Supports lyrics with structure tags and natural-language style descriptions.

Meta MusicGen is an open-source model from Meta that generates music from text prompts or melody conditioning. The melody variant lets you hum or play a tune and generate a full arrangement around it.

Try it out

Test different music models in the playground. Compare outputs side by side to find the right sound for your project.

Open the playground →

Questions? Join us on Discord.