1
Google DeepMind's music model for full songs with vocals, lyrics and SynthID
8.0/10
What it is Lyria is Google DeepMind's music generation family. Lyria 3.5 launched in Google Flow Music on 29 July 2026 and reached the Gemini app, AI Studio and the Gemini API in early September 2026. It writes full-length songs a couple of minutes long, with vocals, timed lyrics and arrangement, from text or image prompts. A separate Lyria 3… Read more →
Pros
- $0.08 per full song via API
- Vocals, lyrics and structure control
- SynthID on all output
- WAV export
Cons
- No free API tier
- Consumer commercial terms unclear
- Few third-party apps so far
Pricing: Paid · text-to-music vocals api synthid google
2
ElevenLabs' music model: songs up to 5 minutes with vocals, cleared for most uses
7.9/10
What it is Eleven Music is ElevenLabs' music generation model. Music v2 launched in May 2026, a month after the ElevenMusic consumer app. Music v2.5 (model ID music_v2_5) reached the API on 14 September 2026 with better audio quality and prompt adherence. ElevenLabs announced a partnership with Universal Music Group on 10 September 2026. Key features - Songs from 3… Read more →
Pros
- Cleared for most commercial uses
- Vocals in several languages
- Section inpainting and fine-tunes
- Same account as ElevenLabs voices
Cons
- Film/TV/game use excluded on self-serve plans
- Free plan not commercial
- Music API needs a paid plan
Pricing: Free plan, paid from $6/mo · text-to-music vocals licensed api elevenlabs
3
Stability AI's licensed-data audio models up to six minutes, with open-weight tiers
★ 0 · Stability AI Community License · archived
7.5/10
What it is Stable Audio is Stability AI's music and sound-effects family. Stable Audio 3.0, announced on 20 May 2026, is a model family trained on fully licensed data. Large is an enterprise model. Medium, for full songs, and Small / Small SFX, for mobile and sound effects, are released as open weights. Stability's investors now include Warner Music Group,… Read more →
Pros
- Trained on fully licensed data
- Open Small and Medium weights
- Up to six-minute tracks
- Fast, runs on a laptop
Cons
- Not built for sung vocals
- Community licence, not OSI open source
- App pricing not verified
Pricing: Open source · text-to-music sound-effects open-weights licensed-data stability-ai
4
MIT-licensed open music model that makes full songs with vocals locally in seconds
★ 13k · MIT · updated 2026-09-03
7.4/10
What it is ACE-Step is an open-source music generation model. ACE-Step 1.5, released on 2 April 2026, generates songs from 10 seconds to 10 minutes long, with vocals and lyrics in 50+ languages. It is published under the MIT licence. An XL variant with a 4B-parameter decoder improves quality for users with bigger GPUs. Key features - Songs from 10… Read more →
Pros
- MIT licence
- Vocals and lyrics in 50+ languages
- Very fast generation
- Runs on modest GPUs
Cons
- Vocals behind Suno/Lyria
- Technical setup
- No hosted consumer app from the makers
Pricing: Open source · open-source text-to-music vocals comfyui local
5
Meta's open research model for short instrumental music from text or a melody
★ 24k · MIT · updated 2026-03-03
6.2/10
What it is MusicGen is Meta's text-to-music model, part of the AudioCraft research library. It generates instrumental music from a text description, optionally guided by a melody you hum or upload. AudioCraft also includes MusicGen Style, MAGNeT and JASCO, a model that takes chords, melodies and drums as conditions. MusicGen was first released in 2023 and is now older than… Read more →
Pros
- Easy to run locally
- Melody conditioning
- Training code included
- MIT-licensed code
Cons
- Weights are non-commercial (CC-BY-NC 4.0)
- No vocals
- Dated quality next to 2026 models
Pricing: Open source · open-source text-to-music research meta non-commercial