
Google Lyria 3 Review (2026) — Clip, Pro, and 3.5
By Olivia Hart
MyClaw Editorial
MyClaw
Run Best-in-Class AI Agents Now
Run OpenClaw or Hermes Agent on managed MyClaw hosting, so your AI agent stays online, updated, and ready for real work.
AI Takeaway
- Is Google’s Lyria worth trying? Yes for hooks, backing music, and song demos. For a track that has to land on exact beats or fit around dialogue, audition it against your edit before committing.
- Which Lyria model should you use? Choose Lyria 3 Clip for a 30-second file and Lyria 3.5 for a new full-song project. Google still lists the older Lyria 3 Pro as an API preview.
- How do you make a full song in Gemini? The app uses Lyria 3.5. Switch Gemini to Pro, open Create music, and select Full for a track of about two to three minutes.
- How much does the Gemini API cost? Clip is $0.04 per generation. Lyria 3 Pro and 3.5 are both $0.08 per song. All three require paid API access.
Lyria 3 Clip, Pro, and 3.5 Compared
Lyria is Google DeepMind’s music generator. It can turn a brief about genre, instrumentation, and mood into a vocal track or an instrumental. Pick a model based on how much music you need and whether the track needs to develop across distinct sections.
| Model | Gemini API model ID | Output and best use | Gemini API price |
|---|---|---|---|
| Lyria 3 Clip | lyria-3-clip-preview | Fixed 30-second clips for hooks, jingles, and arrangement ideas | $0.04 per clip |
| Lyria 3 Pro | lyria-3-pro-preview | Songs up to about three minutes for existing Pro workflows | $0.08 per song |
| Lyria 3.5 | lyria-3.5 | Variable-length songs for new projects, with room for verses, choruses, and bridges | $0.08 per song |
Prices are for the Gemini Developer API in USD, checked November 16, 2026. The Gemini app has its own track limits, which depend on your account and plan.
Clip still earns its place when you want to audition several grooves before paying for full songs. Google lists both Clip and Pro as legacy previews and recommends 3.5 as the replacement for Pro. If you already use Pro, try the listening test from the Suno v6 vs. 5.5 comparison with a Lyria brief you know well. Compare where the instruments come in and how the chorus builds before switching to 3.5.
![]()
The Pro setting in the Gemini app controls access to the full-track option. To target the older Lyria 3 Pro model, use its API model ID.
How Lyria Handles Hooks, Full Songs, and Background Music
DeepMind’s evaluation reported better audio fidelity and closer adherence to lyric prompts for Lyria 3 than for Lyria 2. Google describes 3.5 as an improvement in vocal expression, pronunciation, and arrangement. An independent study by Zoe Slendebroek and Danaé Metaxa found that the earlier Lyria 3 followed requested tempos only moderately and produced less varied audio than human recordings in three of the four genres studied. The study sampled tracks in April and May 2026, before Lyria 3.5 launched.
Short Hooks and Instrumental Cues
Try a brief like “Create a 30-second instrumental for a weekend travel podcast. Open with a bright plucked melody, add brushed drums, and land on a clean final beat.” Check where the hook becomes recognizable and whether the last beat gives you a clean place to cut. If you’re making a teaser with one of the best free AI music video generators, mark a strong beat for the first cut.
For music under narration, try muted chords and a light pulse. Play the cue under the actual voice track at the volume you’ll use. If the music competes with the voice, identify the instrument that crowds it and adjust that part of the brief.
Vocals, Lyrics, and Song Structure
If you’re deciding whether to move from Pro to 3.5, give both models the same original lyrics and label the sections [Verse], [Chorus], and [Bridge]. State if a section should be instrumental. Keep tempo and vocal direction consistent and generate two takes per model to see which traits recur.
Listen at similar volume. Check for skipped lines and unclear words, then listen through the return from the bridge. The chorus may be in the right place and still feel abrupt if the arrangement drops out or builds too early.

What Happens When a Take Needs Changes
If only one passage needs work, the Suno v6 review shows how section editing can help you revise it in place. Each Gemini API request creates a fresh take. Even an identical prompt can yield a different result, so asking for softer brass can also change a chorus you liked.
Where to Generate Lyria Music Today
Gemini for a Quick Track
On the web, open Upload & tools in Gemini and select Create music. Describe the track, choose a genre and whether you want vocals, and attach an image or video if it helps set the mood. The app’s Short option runs about a minute. Switch to Gemini Pro for a full track of about two to three minutes.
If the music tool is missing, check your account. Google’s help page lists a minimum age of 18, a supported signed-in account, and Keep Activity enabled.
If you’re making separate artwork with an AI album cover generator, download the audio as an MP3. Choose MP4 to include Gemini’s generated cover. Gemini embeds Google’s SynthID watermark to help identify AI-generated music.
Flow Music or the API for More Control
Use Google Flow Music to develop a take that already has the right sound. Its section editor can rewrite selected lyrics or restyle a beat drop. Available controls can vary by device, region, and subscription.
Use Google AI Studio to audition a model in the browser or the Gemini API to record the exact model ID and prompt for each generation. For a small API trial, two Pro takes and two 3.5 takes cost $0.32 at the listed rates. Each request counts separately, so budget for the versions you’ll discard as well as the one you keep.
When Lyria Is a Good Fit
- A theme you’ll use more than once. Choose a take with a distinctive opening you can shorten for a trailer or reuse across episodes. Save an uncut copy so you can pull a longer version later.
- A demo you’re sharing for feedback. Ask someone to mark where they expect the chorus and which lyric they remember. Use that feedback to decide whether the next pass needs a different structure or clearer words.
- A soundtrack timed to picture. For API generations, put timestamps for major cues in the prompt. Check those cues against the rough cut. If the title appears at 0:07, test whether you can line up a musical accent there without an audible jump.
Plan and Review Lyria Tracks with MyClaw

A podcast trailer might need a recognizable opening motif and a quieter cue under the host. Keep both briefs and your listening notes together so you can tell which takes are ready for the edit.
1. Turn the project into a music brief
Draft the two briefs in MyClaw Chat with the length of each cue, the point where the host starts speaking, and the instruments you have in mind. Ask it to write two versions of the brief that vary one element, such as plucked guitar versus piano, so you can hear what that change does to the feel.
2. Audition your Google-generated takes
Generate the tracks in Gemini or through the Gemini API. Bring the filenames, durations, and your listening notes back to the chat. For example, “piano take 02 leaves the voice clear but builds at 0:12” gives you enough detail to ask for a prompt that delays the build until the host finishes.
3. Keep your cue list ready for the next release
If you publish regularly, a private, always-on OpenClaw agent hosted by MyClaw can keep a cue log with the model, prompt, chosen file, and review status. Schedule a reminder before your next edit with the takes that still need a decision.
Conclusion
Before budgeting for a batch of tracks, take one promising file all the way through your edit. If a trim and a fade get it ready, save the prompt and model so you can use the same brief as a starting point next time. If a small change sends you back through several generations, factor in time for section editing when you plan the rest of the project.
Skip the Setup, Run Best-in-Class AI Agents Now
Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.