AI music generation is moving fast in 2026, but a polished demo does not tell you which model will handle your song best. This comparison puts ElevenLabs Music v2, Google Lyria 3 Pro, and Mureka v9 through the same three creative briefs: a vocal pop song, a cinematic instrumental, and a structure-control challenge.
All three music models are available through GlobalGPT’s multi-model platform. GPT created the two original lyric sets once, and we kept them unchanged across the music models. Results are separated into performance and efficiency so musical quality is not confused with request cost or generation time.
Quick Answer: How the Three Models Performed
- Prestaties: ElevenLabs led lyric and structure control; Lyria matched the cinematic brief most closely.
- Efficiency: ElevenLabs was fastest, while Mureka had the lowest task-reported charge and the longest outputs.

Model Overview: ElevenLabs Music v2, Google Lyria 3 Pro, and Mureka v9
| Model | Ontwikkelaar | Product focus | Formats relevant to this comparison | Official product page |
|---|---|---|---|---|
| ElevenLabs Music v2 | ElevenLabs | Text-directed music and full-song generation | Music with vocals or instrumental output | ElevenLabs Music |
| Google Lyria 3 Pro | Google DeepMind | Music generation from a written creative direction | Vocal and instrumental briefs entered through one prompt | Google DeepMind Lyria |
| Mureka v9 | Mureka | AI song creation and background-music generation | Separate song and BGM routes in the tested API | Mureka |
The product-family descriptions above come from the developers’ public product pages. Exact request fields used in this article come from the Broly API schema checked on August 5, 2026.
What Is ElevenLabs Music v2?
- Generates complete music from a written brief.
- The tested route accepts creative direction and lyrics in one prompt.
- Unique test control: explicit duration and instrumental-only settings.

What Is Google Lyria 3 Pro?
- The higher-quality Lyria 3 route selected for this comparison.
- The tested schema accepts one combined music prompt.
- Control challenge: no separate lyric, style, or duration fields in this route.

What Is Mureka v9?
- Uses separate Song and BGM routes.
- Song requests separate exact lyrics from style direction.
- Route used: BGM for Test 2; Song for Tests 1 and 3.

Same-Prompt Music Tests
How we test: The same creative requirements go to all three models, with only the field mapping changed for each API.
- Shared inputs: identical briefs and identical GPT-written lyrics from GlobalGPT.
- Duur: every brief targets 90 seconds; only ElevenLabs exposes a duration field.
- Efficiency: measured from submission to terminal status, including polling and network time.
- Listening review: editor notes cross-checked with two-pass Gemini 3.6 Flash audio analysis; unreliable exact timestamps excluded.
Test 01Contemporary Pop Song With Vocals
Opzet van de taak: Create a contemporary synth-pop song about two people leaving a city late at night and moving toward an uncertain future. The evaluation focuses on vocal naturalness, lyric preservation, hook clarity, arrangement, and whether the emotional lift reaches the chorus.
- ElevenLabs and Lyria had no clear lyric deviation across the checks.
- ElevenLabs paired that consistency with the cleanest vocal/instrument separation.
- Mureka sounded distinctive, but its blind transcript flagged likely substitutions.
ElevenLabs Music v2
Test output
- No clear lyric deviation detected.
- Electronic synth-pop character with a noticeable intro.
- Clean vocal and instrument separation.
Google Lyria 3 Pro
Test output
- No clear lyric deviation detected.
- Broad, theme-song-like lift with immediate vocal entry.
- Strong finish, but the ending cuts off abruptly.
Mureka v9
Test output
- Blind transcript flagged a likely lyric substitution.
- Distinctive electronic, nu-disco-like character.
- Extended instrumental outro made it much longer.
Test 02Cinematic Instrumental Music
Opzet van de taak: Build a cinematic cue with an audible narrative arc rather than a static loop. The evaluation focuses on motif development, orchestration, dynamic control, transitions, and whether the final section earns its larger scale.
- Lyria preserved the piano motif, gradual layering, late brass, and resolved ending.
- ElevenLabs followed the arc closely but introduced an impact earlier than requested.
- Mureka delivered a compelling different composition, with voice-like texture that broke the brief.
ElevenLabs Music v2
Test output
- No voice texture detected.
- Recognizable piano motif and controlled build.
- An early impact weakened strict compliance.
Google Lyria 3 Pro
Test output
- Best match for motif, layers, late brass, and climax.
- No voice texture detected.
- Natural movement and a resolved ending.
Mureka v9 BGM
Test output
- Voice-like “ah” textures broke the no-voice brief.
- Violin-led, mournful opening grew into a late climax.
- Distinctive, but several requested elements were replaced.
Test 03Song Structure and Prompt Control
Opzet van de taak: Create an alternative-pop song about repairing an old radio and hearing a distant reply through the static. The required order is Intro, Verse, Pre-Chorus, Chorus, Instrumental Break, Bridge, Final Chorus, and Outro. The evaluation focuses on structural compliance, lyric preservation, contrast between sections, and ending control.
- ElevenLabs made the eight requested sections easiest to identify.
- Lyria created good section contrast but changed more supplied words.
- Mureka kept a polished groove, yet the bridge and outro were less distinct.
ElevenLabs Music v2
Test output
- Clearest eight-part structure in the test.
- Strong bridge and final-chorus contrast.
- Break ran long; ending behaved more like a fade.
Google Lyria 3 Pro
Test output
- Good eight-part flow and near-eight-second break.
- Strong bridge-to-final-chorus contrast.
- Blind transcript flagged changed early lyrics.
Mureka v9
Test output
- Clear vocal hook and polished pop-funk production.
- Less distinct section contrast through bridge and outro.
- Extended guitar outro created the highest edit burden.
What the Tests Reveal About Each Capability
Most consistent supplied words and clearest eight-part structure.
Closest match to the requested motif, layers, brass timing, and ending.
Most stylistically independent outputs, with the greatest edit burden.
Task-level findings from three prompts—not a universal model ranking or speech-accuracy benchmark.
Generation Efficiency
- Complete test pack
9 of 9 usable outputs - Total task-reported charge
$1.05 · no music task rerun
Each bar group uses its own scale. Charges come from completed Broly task responses on August 5, 2026—not official provider pricing or GlobalGPT credits. Local time includes polling and network delay.
Comparing the Three Models Inside GlobalGPT
GlobalGPT keeps the lyric-writing and music-generation stages in one multi-model platform:
- One place to compare: ElevenLabs Music v2, Google Lyria 3 Pro, Mureka v9, and GPT.
- Shared test inputs: the same creative requirements, lyrics, and judging criteria.
- Fair field mapping: each brief is translated into the fields its model route actually accepts.
Want to compare more than one music model before committing?
Explore AI models in GlobalGPT
Start with the same brief, listen for the differences that matter to your project, and keep the output that fits.
Veelgestelde vragen
What is the best AI music generator in 2026?
There is no universal winner from three prompts. In this test pack, ElevenLabs was strongest for lyric and structure control, Lyria followed the cinematic instrumental brief most closely, and Mureka offered the lowest task-reported charge and longest, most stylistically independent outputs.
Can these AI music generators create both vocals and instrumentals?
Yes, but the routes differ. ElevenLabs Music v2 supports vocal and instrumental requests, Lyria 3 Pro receives both kinds of briefs through one prompt, and Mureka v9 uses separate song and BGM routes in this test.
Did every model receive the same prompt?
The creative brief and lyrics are identical, but the request payloads are not. We map the shared instructions to each API’s required fields because ElevenLabs, Lyria, and Mureka expose different schemas.
Are the songs exactly the same length?
No. ElevenLabs returned 90.0 seconds in all three tests, Lyria returned 84.0 to 89.5 seconds, and Mureka returned 167.1 to 183.9 seconds. Only the tested ElevenLabs route exposed a duration field.
Is the listed request cost the same as a subscription price?
No. The efficiency section uses the charge exposed by each completed Broly task response. It does not represent an official provider subscription, an official provider API rate, or a conversion from GlobalGPT credits.
Can I use AI-generated music commercially?
Commercial-use rights depend on the service, account, plan, output, and current terms. Review the applicable provider and platform terms for your project; this creative test does not provide legal clearance.
Are all three models available in GlobalGPT?
Yes. ElevenLabs Music v2, Google Lyria 3 Pro, and Mureka v9 are available through GlobalGPT, allowing users to access multiple music models from one platform.
Eindoordeel
- ElevenLabs Music v2: strongest controlFastest average run, exact 90-second outputs, best lyric consistency, and clearest fixed structure. Tradeoff: highest task-reported charge.
- Google Lyria 3 Pro: strongest cinematic matchBest instrumental brief alignment and close-to-target duration. Tradeoff: abrupt Test 1 ending and lyric changes in Test 3.
- Mureka v9: most independent interpretationDistinctive long-form tracks and lowest task-reported charge. Tradeoff: longest outputs and highest edit burden under strict briefs.



