Availability checked: 7 September 2026. Google announced Lyria 3.5 on 4 September 2026. Its announcement says the model is available in the Gemini app and Gemini API, with the Gemini app available globally on web and mobile. Google also names Flow Music, Google AI Studio and Google Vids as additional surfaces. Product access, limits and terms can change, so treat this as a dated availability record and recheck the linked pages before a client launch.
Lyria 3.5 is useful to marketers because it turns an audio brief into something a team can hear and review early. That does not make an output a cleared campaign asset. The practical job is to use the model for exploration, make provenance easy to audit, and route every commercial use through rights and policy review.
Where Lyria 3.5 is available
| Surface | What the official announcement says | Marketer use | Check before work |
|---|---|---|---|
| Gemini app | Available globally on the web and mobile app. | Prompt a concept, backing track, jingle or short/long variation. | Account access, plan, output controls and current terms. |
| Gemini API | Lyria 3.5 is available through the API. | Prototype an internal audio workflow or controlled batch of variants. | API account, quota, billing, retention and permitted use. |
| Flow Music | Google names Flow Music as a surface for artists and AI creatives. | Explore arrangements and references with a musician or creative lead. | Product terms, export path and rights documentation. |
| Google AI Studio | Google names AI Studio for developers and technologists. | Test prompts and build a reproducible technical proof of concept. | Model availability, API settings and account controls. |
| Google Vids | Google names Vids as another place to use Lyria 3.5. | Develop a video soundtrack concept beside the visual storyboard. | Workspace or account rollout and downstream asset permissions. |
Read the Google Lyria 3.5 announcement for the launch scope. It explicitly gives a custom backing track, a brand jingle and a personalized ringtone as examples. Those examples explain the reader job; they do not grant a commercial licence.
What the model can produce
Google says Gemini users can select or describe a genre, choose vocal or instrumental styles, use templates and choose short or longer tracks. The Google DeepMind Lyria page describes cohesive tracks up to three minutes, image-to-music, and imperceptible SynthID watermarking. It also says users can direct musical details and create high-fidelity audio. Treat those as documented capabilities, not as a promise of a particular mix, pronunciation, vocal identity, file format, platform performance or audience response.
For a video team, the most useful distinction is between a sketch and a release asset. A 30-second concept can help a creative lead decide whether a storyboard needs a warm acoustic bed, a high-energy electronic pulse or a restrained instrumental. A longer generation can reveal whether a theme survives beyond the opening idea. Both outputs still need listening tests, editing and rights review.
Practical marketer workflows
- Storyboard soundtrack exploration. Write the scene, duration, mood, tempo range, instrumentation and edit points. Generate a few clearly labeled concepts, then let the creative owner choose a direction. A related DMT video-generation guide covers the visual production side; keep the audio provenance beside the video brief.
- Brand-jingle discovery. Use Google’s jingle example as an ideation brief. Define what the brand can and cannot sound like, avoid naming a living artist as a style shortcut, and save the prompt and output with the review record. A human composer or rights reviewer should decide whether any motif is usable.
- Internal mood-board audio. Use image-to-music to turn an approved visual reference into a private concept track. Record the source image, who supplied it, the account used and the date. Do not place confidential client material into a generation workflow until the account and data policy permit it.
- API proof of concept. If a developer tests variants through the Gemini API, store the model/surface information exposed by the account, request parameters, output identifiers, costs and retention decision. Do not assume that a successful API response settles distribution or advertising rights.
Rights, safety and provenance
The Lyria 3.5 model card, updated in July 2026, describes the model, distribution and safety approach. It says Google’s Generative AI Prohibited Use Policy applies and that uses violating privacy or intellectual-property rights are not acceptable. It also describes product-level safety filtering and SynthID. The SynthID overview explains that the audio watermark is intended to be inaudible and detectable by Google’s technology.
A watermark helps with provenance; it is not a rights clearance certificate. Google’s availability announcement also does not say that every generated melody, lyric, voice-like performance, image input or brand reference is cleared for paid advertising, client delivery or worldwide distribution. Before use, read the current terms for the exact surface, confirm whether commercial use is allowed, check the territory and contract, and obtain approval for third-party lyrics, voices, recordings, samples, images and recognizable references.
Pre-campaign checklist
- Record the Google surface, account, plan or API project and date.
- Keep the brief, prompt, source image, output file and any model or output identifier together.
- Listen for unintended resemblance, unsafe content, unsupported claims, lyrics or voice characteristics.
- Check the current Gemini, API or Flow terms for commercial use, redistribution, attribution and retention.
- Preserve the SynthID or other provenance signal through the approved edit and export path where possible.
- Obtain creative, legal, client and platform approval before placing the audio in paid media or a public campaign.
Editorial boundary: Lyria 3.5 is a documented option for audio exploration across Google surfaces. This article does not promise commercial-use rights, originality, approval, performance or availability for a particular account. Recheck the official pages and the account-level terms immediately before publication or distribution.
