1M-3MPublicB2CB2D (developers)Product launchGlobalDemoUSVertical AI
TLVC Rating
3.60
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Lyria 3.5 is Google's text-to-music model, now available inside Google AI Studio, through the Gemini API for developers, and in the consumer Gemini app. It generates full songs from a written prompt, with tracks running up to about three minutes and improvements to vocals, lyric adherence, and accent handling . This launch matters because Google is putting the same model into a consumer chat surface and a developer API simultaneously, rather than staging the usual months-long gap between the two.
For builders, the Gemini API entry means Lyria 3.5 can be wired into apps that need original soundtracks, background beds for video, or interactive audio without leaning on stock libraries. The model first appeared inside Google Flow Music, and five weeks later arrived across the Gemini app, the Gemini API, and Google AI Studio for prototyping , giving product teams a first-party option that already sits next to Gemini, Veo, and Imagen in the same tooling.
The model comes from Google DeepMind, the group behind the broader Gemini family. For founders and operators comparing AI music tools like Suno and Udio, the practical read is that Lyria 3.5 is reachable from the same API key and console you likely already use for text and multimodal work, which lowers the cost of adding generated music to an existing product.
Comments (15)
Sign in to join the discussion.
N
Nadia RamirezSep 6, 2026
The demo track in the announcement was suspiciously well-mixed. Curious how much of that fidelity survives when a random user types 'lofi but sad' at 2am.
K
Kenji MatsuokaSep 6, 2026
Expressive vocals is doing a lot of heavy lifting in that sentence. Show me a rap verse with actual consonants and I'll believe it.
H
Henrik BjornlundSep 6, 2026
Was the launch clip rendered in real time or batched? The transitions between stems feel too clean for a live sample.
P
Priya VenkateshSep 6, 2026
Before anyone ships this into a product: who owns the output, and does training data provenance get disclosed in the API response? Asking for every legal team on earth.
O
Oleksiy TkachenkoSep 6, 2026
The tweet buries the actual news under three deployment surfaces. Pick a lane, the copy reads like a release note approved by four VPs.
M
Marisol QuinteroSep 6, 2026
Tapped through AI Studio and hit the music tab in about 12 seconds. Onboarding is fine, but the prompt box gives zero hints about what actually works well.
T
Tanvi R.Sep 6, 2026
New PM here, genuine question: if I generate a song in the Gemini app, can I actually download the stems, or just the final mix? The blog isn't clear.
D
Devon OkaforSep 6, 2026
Every generative music launch since 2022 has claimed 'richer arrangements.' At some point richer has to mean something measurable.
Y
Yuki T.Sep 6, 2026
The Gemini app rollout is the interesting bit here, not the model card. Distribution is the moat, everyone else is fighting for a Discord bot slot.
L
Leo FeinbergSep 6, 2026
Cool tech, but who's paying for AI-generated songs at scale? Sync licensing is a knife fight and consumer retention on novelty music tools is brutal.
A
amaranthSep 6, 2026
Docs page for the Lyria endpoint is thin. Where are the sample rates, latency numbers, and streaming support? Devrel, do your thing.
R
Roshan PillaiSep 6, 2026
Any paper or tech report accompanying 3.5, or are we just supposed to vibe-check the outputs against 3.0? Would love architectural notes.
C
Cassie WallenbergSep 6, 2026
Politely following up: is there a watermarking scheme on Lyria outputs, and can platforms detect it reliably? Happy to take this to email.
B
Bruno SantosSep 6, 2026
We shipped basically this inside a Google music prototype in 2019 and it never left the demo phase. Glad it finally found a launch tweet.
M
Maksym OrlovSep 6, 2026
Inevitable take: someone will wrap this in a token-gated 'AI record label' within a week. Please don't.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Google
How we're making AI helpful for everyone
How we're making AI helpful for everyone
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Lyria 3.5 is Google's text-to-music model, now available inside Google AI Studio, through the Gemini API for developers, and in the consumer Gemini app. It generates full songs from a written prompt, with tracks running up to about three minutes and improvements to vocals, lyric adherence, and accent handling . This launch matters because Google is putting the same model into a consumer chat surface and a developer API simultaneously, rather than staging the usual months-long gap between the two. For builders, the Gemini API entry means Lyria 3.5 can be wired into apps that need original soundtracks, background beds for video, or interactive audio without leaning on stock libraries. The model first appeared inside Google Flow Music, and five weeks later arrived across the Gemini app, the Gemini API, and Google AI Studio for prototyping , giving product teams a first-party option that already sits next to Gemini, Veo, and Imagen in the same tooling. The model comes from Google DeepMind, the group behind the broader Gemini family. For founders and operators comparing AI music tools like Suno and Udio, the practical read is that Lyria 3.5 is reachable from the same API key and console you likely already use for text and multimodal work, which lowers the cost of adding generated music to an existing product.
Comments (15)
The demo track in the announcement was suspiciously well-mixed. Curious how much of that fidelity survives when a random user types 'lofi but sad' at 2am.
Expressive vocals is doing a lot of heavy lifting in that sentence. Show me a rap verse with actual consonants and I'll believe it.
Was the launch clip rendered in real time or batched? The transitions between stems feel too clean for a live sample.
Before anyone ships this into a product: who owns the output, and does training data provenance get disclosed in the API response? Asking for every legal team on earth.
The tweet buries the actual news under three deployment surfaces. Pick a lane, the copy reads like a release note approved by four VPs.
Tapped through AI Studio and hit the music tab in about 12 seconds. Onboarding is fine, but the prompt box gives zero hints about what actually works well.
New PM here, genuine question: if I generate a song in the Gemini app, can I actually download the stems, or just the final mix? The blog isn't clear.
Every generative music launch since 2022 has claimed 'richer arrangements.' At some point richer has to mean something measurable.
The Gemini app rollout is the interesting bit here, not the model card. Distribution is the moat, everyone else is fighting for a Discord bot slot.
Cool tech, but who's paying for AI-generated songs at scale? Sync licensing is a knife fight and consumer retention on novelty music tools is brutal.
Docs page for the Lyria endpoint is thin. Where are the sample rates, latency numbers, and streaming support? Devrel, do your thing.
Any paper or tech report accompanying 3.5, or are we just supposed to vibe-check the outputs against 3.0? Would love architectural notes.
Politely following up: is there a watermarking scheme on Lyria outputs, and can platforms detect it reliably? Happy to take this to email.
We shipped basically this inside a Google music prototype in 2019 and it never left the demo phase. Glad it finally found a launch tweet.
Inevitable take: someone will wrap this in a token-gated 'AI record label' within a week. Please don't.