Music GenerationActive
ElevenLabs Text to Music v2.5 Lyrics
Generate full songs with vocals from a structured composition plan. Control section timing, lyrics, style keywords, and MP3 output quality with ElevenLabs Music v2.5.
Overview
Text to Music v2.5 Lyrics is an audio generation model by ElevenLabs. It turns a structured composition plan into a complete, mixed song with vocals and instrumentation. You define the song as timed sections, then the model performs the lyrics and arranges the backing track to match. This format solves the common prompt problem of losing structure, since timing and section boundaries are explicit.
What you can build
- Full songs with verses and choruses that land at specific times
- Branded jingles with repeatable structure across multiple variants
- Podcast and video intro themes with sung taglines
- Game or trailer cues with controlled transitions between sections
- Multilingual vocal demos where you provide the lyrics to sing
Inputs
- A composition plan provided as a JSON object that contains a list of up to 30 song sections.
- For each section, you provide a bracketed section label (for example, Verse or Chorus) plus the lyrics as plain text. Put each lyric line on its own new line. You can leave the lyrics empty for an instrumental-only section.
- For each section, you set an exact section duration in milliseconds. Each section must be between 3,000 ms and 120,000 ms.
- For each section, you add style keywords to include and style keywords to avoid. Each list supports up to 50 entries. Write style keywords in English.
- Optionally, you can set how strongly a section should stay consistent with nearby sections. This helps keep themes stable across the whole track.
- An MP3 output quality choice that sets the sample rate and bitrate for the returned track.
Outputs
The model returns a single MP3 file.
It contains the full stereo mix for the whole song. Vocals and instruments come merged as one finished track. The audio sample rate and bitrate match the output quality you selected.
Recommended settings
- Put 6 to 7 clear style keywords in the first section. That first section sets the overall genre and mood.
- Include tempo as a style keyword when you care about pacing. For example, include a BPM value.
- Use “avoid” style keywords to block unwanted artifacts. For example, add unwanted vocal processing or instruments.
- Use high continuity between sections when you want recurring motifs and consistent vocal character.
Limitations
- The song must be between 3 seconds and 10 minutes total.
- Each section must be 3 to 120 seconds. Longer passages must be split into multiple sections.
- The model follows section durations strictly for Music v2.5. If timings are awkward, phrasing may sound forced.
- Style keywords must be in English, even when the lyrics are not.
- Messy inputs can break results. Invalid JSON, inconsistent section labels, or very long unbroken lyric lines can reduce quality.
- Output is a single mixed track. It does not return stems or MIDI in this model run.
Safety & compliance
- Don’t include artist names, songwriter names, song titles, album titles, label names, or publisher names in your input.
- Don’t paste substantial or distinctive lyrics from existing songs.
- If your styles include copyrighted references, the request can fail with a composition-plan validation error and a suggested safer alternative.
- Review the generated audio before publishing. You are responsible for rights clearance and required disclosures for your use case.
Example prompts
Great starting points for text to music v2.5 lyrics.
{"chunks":[{"text":"[Verse]\nMirror ball is spinning round\nFeel the bass beneath the ground","duration_ms":10000,"positive_styles":["70s disco funk","slap bass","four on the floor drums","string stabs","soulful female vocals","118 bpm","great production quality"],"negative_styles":["distorted guitars","autotune"]},{"text":"[Chorus]\nDance with me tonight\nUnder golden light","duration_ms":10000,"positive_styles":["catchy disco chorus","layered harmonies","bright horns"]}]}Music Generation
{"chunks":[{"text":"[Verse]\nClose your eyes, the stars are near\nMoonlight sings for you, my dear","duration_ms":10000,"positive_styles":["gentle lullaby","soft fingerpicked acoustic guitar","music box","tender female vocals","65 bpm","warm and intimate","great production quality"],"negative_styles":["drums","electric guitars"]},{"text":"[Chorus]\nSleep now, sleep now\nDreams will show you how","duration_ms":10000,"positive_styles":["soft humming harmonies","light strings"]}]}Music Generation
{"chunks":[{"text":"[Verse]\nWoke up this morning with the rain on my door\nMy old guitar is all I am living for","duration_ms":10000,"positive_styles":["gritty blues rock","overdriven electric guitar riff","shuffle drums","raspy male vocals","95 bpm","great production quality"],"negative_styles":["synths","pop"]},{"text":"[Chorus]\nRoll on, river, roll on\nCarry my troubles till they are gone","duration_ms":10000,"positive_styles":["wailing guitar fills","stomping groove"]}]}Music Generation
{"chunks":[{"text":"[Build]\nHands up high, we are touching the sky","duration_ms":10000,"positive_styles":["energetic edm pop","rising synth build","snare roll","bright female vocals","128 bpm","great production quality"],"negative_styles":["acoustic","slow tempo"]},{"text":"[Drop]\nOh oh, we are alive","duration_ms":10000,"positive_styles":["big festival drop","punchy supersaw synths","heavy sidechained bass"]}]}Music Generation
API quick start
Run text to music v2.5 lyrics with a single API call.
{
"prompt": "{\"chunks\":[{\"text\":\"[Verse]\\nMirror ball …",
"outputFormat": "mp3_44100_128"
}curl -X POST "https://api.wiro.ai/v1/Run/elevenlabs/text-to-music-v2-5-lyrics" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "{\"chunks\":[{\"text\":\"[Verse]\\nMirror ball …",
"outputFormat": "mp3_44100_128"
}
JSONOther models you might like
Explore similar music generation models.
