Lyria 3.5 Workflow: 30-Second Gemini Clips, Then Full Songs, Custom Lyrics, and Video Hooks
Lyria 3.5 is in Gemini and the Gemini API. Draft 30-second clips first, then generate full songs with custom lyrics, and turn the approved hook into video.
Lyria 3.5 Workflow: 30-Second Gemini Clips, Then Full Songs, Custom Lyrics, and Video Hooks
Lyria 3.5 is Google’s current music model, not a video generator and not a one-click “make a song” button. Google’s 4 September 2026 Gemini post says Lyria 3.5 is now in the Gemini app and the Gemini API, with more expressive vocals and richer arrangements. The Gemini API music docs (updated the same day) split the family: Lyria 3 Clip (lyria-3-clip-preview) always returns a 30-second MP3; Lyria 3.5 (lyria-3.5) generates a full song that lasts a couple of minutes, with duration steered by the prompt or timestamps. ClipCanva is not affiliated with Google, Canva, or Suno, and ClipCanva does not host the Lyria identifiers. Draft the 30-second clip on Gemini or the API, lock lyrics on ClipCanva’s AI lyrics generator, then turn the approved hook into scripts and clips on the AI script generator and AI video generator.
Official facts for Lyria 3.5 and nearby desks
| Source | Public claim | Workflow implication |
|---|---|---|
| Create your best tracks yet with Lyria 3.5 in Gemini (4 Sep 2026) | Lyria 3.5 is in the Gemini app and Gemini API. In Gemini you can pick or describe a genre, choose vocal or instrumental, use templates, and choose short or longer tracks. Google lists backing tracks, brand jingles, and ringtones as jobs. Global on web and mobile. Also in Google Flow Music, Google AI Studio, and Google Vids. | The 4 September change is distribution, not a new model family. Confirm whether your job is a 30-second clip or a full song before you generate. |
| Gemini API: Generate music with Lyria 3.5 (updated 4 Sep 2026) | Clip: lyria-3-clip-preview, 30 seconds, MP3. Lyria 3.5: lyria-3.5, a couple of minutes, MP3 by default, WAV available via response_format. Output is 44.1 kHz stereo. Up to 10 images with the text prompt. Custom lyrics with [Verse] / [Chorus] / [Bridge]. Timestamp blocks for structure. Prompt language drives lyric language. |
Iterate on Clip. Pay for a full song only after genre, BPM, and hook survive. Do not send ten product stills to Clip if you only needed a 30-second bed. |
| Same API docs — limitations | Safety filters block artist-voice and copyrighted-lyric requests. All audio includes a SynthID watermark. Single-turn: no multi-turn edit of a generated clip. Clip length is fixed at 30 seconds. Full-song length is “a couple of minutes,” influenced by the prompt. Same prompt can yield different takes. | Lyria 3.5 is not ElevenMusic Composer. If verse one is a keeper and chorus two is wrong, you cannot currently rewrite only the chorus through the API. |
| DeepMind: Lyria | Positions Lyria as Google’s most advanced music model, up to 3 minutes. Prompt for a 60-second clip, a half-length track, or a full 3-minute song. Compose from an uploaded image. | Treat “up to 3 minutes” as the product-page ceiling. The API still says “a couple of minutes.” Time the live output; do not promise 180 seconds from a vague prompt. |
| Introducing Lyria 3.5 in Google Flow Music (29 Jul 2026) | Flow Music launch: richer melody, better lyric prompt adherence and structure, more expressive vocals, plus tempo and duration control. | Flow Music is the artist desk. Gemini is the consumer desk. The API is the developer desk. Do not mix their controls in one brief. |
| Canva AI Music Generator | Style tags, duration, intensity. Free: 900 tokens/month, up to 10 soundtracks/day. Maximum generated length 180 seconds. Free plan: personal use, no standalone download. Paid: commercial use inside Canva designs/exports; raw audio still cannot be downloaded separately. Licensing is on Canva’s music partner terms. | An 8-second Canva layout clip and a 30-second Lyria Clip are different files. Do not treat a Canva soundtrack as a master you can stem later. |
Region, plan gates, safety filters, and which surface your account actually exposes can change. Confirm the live Gemini, API, Flow Music, or Canva panel before you quote 3 minutes, WAV, or image-to-music.
Pick the desk before you generate a full song
Lyria 3.5 fails when every brief is “make me a song.” The 4 September split is four different jobs.
| Job | Where it actually runs | What you must supply | What the model may invent | Fail condition |
|---|---|---|---|---|
| Twelve cheap beds for a 15-second Reel | Gemini or API Lyria 3 Clip | Genre, BPM, vocal vs instrumental, 30s is the output even if the video is 15s | Melody, drums, any vocal you did not forbid | Asking Clip for a 2-minute verse-chorus song |
| Full track with verses and a bridge | API lyria-3.5 or Gemini “longer tracks” / Flow Music |
Locked lyrics or a timestamp map, plus duration in the prompt | Arrangement, pronunciation, extra bars | Sending the first Clip take straight to video |
| Product stills → mood bed | Lyria 3.5 with up to 10 images plus a text prompt | Approved stills that share lighting and identity | Instruments and tempo unless you lock them | Ten unrelated screenshots and no BPM |
| Custom sung lines | API custom-lyrics prompt with [Verse] / [Chorus] |
Verbatim lines you are allowed to ship | Melody and phrasing of those lines | Artist-clone or copyrighted lyric requests (blocked) |
| One bad chorus, keep the verse | Not Lyria API today | n/a | n/a | Expecting multi-turn section edits; that is Composer, not Lyria |
| Layout soundtrack inside a design file | Canva AI Music Generator | Mood tag, duration ≤ 180s | A bed that stays inside Canva | Needing a standalone WAV master |
| Hook → Short or ad | ClipCanva lyrics, script, video | The approved chorus, not the discarded Clip | A new visual and on-screen caption | Animating lyrics you already replaced |
If the brief is “does this BPM even work under a product demo,” that is Clip. If the brief is “this chorus must be sung as written,” that is Lyria 3.5 with custom lyrics. If the brief is “verse one is keepers,” leave Lyria and use a section editor.
Clip vs Lyria 3.5 vs a Canva soundtrack
| Dimension | Lyria 3 Clip | Lyria 3.5 | Canva AI Music Generator |
|---|---|---|---|
| Official ID / surface | lyria-3-clip-preview; Gemini short tracks |
lyria-3.5; Gemini longer tracks, Flow Music, API |
Canva Video Editor feature (partner-powered) |
| Duration | Always 30 seconds | A couple of minutes; DeepMind page says up to 3 minutes | Up to 180 seconds |
| Default file | MP3, 44.1 kHz stereo | MP3; WAV via response_format |
Stays inside the Canva design unless your plan allows export of the project |
| Images | Not the Clip-first path in Google’s best-practice note | Up to 10 images with text | Not documented as image-to-music on the Canva music page |
| Custom lyrics | Possible in a 30s form; still a single turn | [Verse] / [Chorus] / [Bridge] plus timestamps |
Canva’s separate lyric generator is Magic Write text, not this soundtrack tool |
| Edits after generate | Single-turn; regenerate the whole clip | Single-turn; regenerate the whole song | Tweak length and intensity inside Canva |
| Use it for | Prompt tests, loops, social beds | Keepers, jingles, 2–3 minute cues | A bed that never leaves the Canva file |
Do not mix the three in one prompt. A Clip that “sounds good enough at 30 seconds” is a candidate, not a master. Stretching it to two minutes inside a video editor is not the same as generating lyria-3.5.
Draft-then-keeper workflow
Write the packet once. Change only the variable. Promote only winners.
- Name the output. Example: twelve 15-second 9:16 product Reels, same bottle, four cameras, one 8-bar hook, no sung brand claims.
- Lock the lyric spine first. Theme, POV, forbidden claims, hook length. Draft on the AI lyrics generator. Mark
[Chorus]as the only lines that may appear on screen. - Run Clip at 30 seconds. Genre, BPM, key, vocal or instrumental. Google’s API note is explicit: iterate with Clip before a full-length
lyria-3.5call. - Score the test set. Hook memorability, lyric audibility, no invented artist name, no trademarked slogan. Kill the rest.
- Promote one packet to Lyria 3.5. Same lyrics, same BPM, duration in the prompt (
90-second cueor timestamp map). Do not “improve” the chorus while you change length. - Human-gate the vocal. Confirm the sung line matches the approved text. SynthID is present whether you hear it or not.
- Hand the hook to video. Script the 15-second visual on the AI script generator. Generate picture on the AI video generator or image to video. If you need a music-video cut, use AI music video against the approved chorus, not the discarded Clip.
For stills that already exist, attach them only on the keeper pass. The API allows up to 10 images; that is a mood brief, not a product-label guarantee.
Operator card for a 12-clip SKU bed
Job: 12x 15s 9:16 product Reels, same bottle, one 8-bar hook
Draft desk: Gemini / API lyria-3-clip-preview — 30s MP3, instrumental unless the hook is sung
Keeper desk: lyria-3.5 — 90s cue, same BPM, custom lyrics if sung
Locked: SKU name spoken only if legal approved; no competitor names; no artist-clone
Variable: BPM (100 / 118 / 128) and bed type (pluck / lo-fi / short brass sting)
Lyric packet:
[Chorus]
Open the cap. Hear the fizz. Stay for the pour.
(original; not a vendor sample)
Clip prompt pattern:
30-second instrumental, [genre], [BPM], [key], no vocals, no lyrics.
Tight drums, short tail, safe for a 15s cut at bar 1 or bar 5.
Keeper prompt pattern:
90-second [genre] cue at [BPM] in [key].
[0:00-0:08] sting
[0:08-0:45] verse bed under VO
[0:45-1:10] chorus energy
[1:10-1:30] fade for end card
Lyrics: only the approved [Chorus] if vocals are on
Test set: 3 Clip takes (one BPM) before any lyria-3.5 call
Promote: 1–2 winners to full song; do not rewrite the chorus on the keeper pass
Forbidden: copyrighted lyrics, artist-voice requests, Canva 180s beds as the master
Video handoff: ClipCanva script + video generator; captions from the approved chorus only
Example, podcast bumper: lock a 6-word spoken ID as lyrics; run three instrumental Clips at 90 / 100 / 110 BPM; pick the bed that does not mask the ID; then generate a 60-second lyria-3.5 bumper with timestamps. Do not start with a 3-minute song.
Nearby tools, not substitutes
Suno’s Studio 2.0 (13 August 2026) is a browser DAW: MIDI, stems, mix. That is post-Lyria work. ElevenMusic Composer is section-by-section rewrite; Lyria’s API is single-turn. Canva scores a design file. ClipCanva’s AI music generator is the in-house prompt desk when you need a reviewable track without switching to Gemini. None of these are Google Lyria, and none of them replace a legal review of vocals, brand names, or ad claims.
If you already keep a lyrics-to-video packet, the earlier notes still apply: Suno vs Udio vs Canva for provider choice, and ElevenMusic Composer when you must edit one section without regenerating the song. Lyria 3.5 sits in front of those pages as the 30-second draft desk that just landed in Gemini.
Creator checklist
- Write the job as Clip (30s) or full song (prompted minutes). Do not leave duration implied.
- Separate lyrics from production notes. Google’s docs ask for that split.
- Prompt in the language you want sung.
- Forbid artist names, copyrighted lines, and unapproved product claims in the prompt.
- Generate three Clip takes before one
lyria-3.5call. - Listen on phone speakers. If the hook dies, the 15-second Reel dies.
- Export the keeper, then script the visual from the same chorus text.
- Keep SynthID and platform terms in the delivery notes. Do not strip watermarks.
- Confirm commercial use on the surface you actually used (Gemini, API, Flow Music, Canva, ClipCanva). Terms differ.
FAQ
Is Lyria 3.5 free in Gemini?
Google’s 4 September post says Lyria 3.5 is available to all users globally in the Gemini web and mobile apps, and also in Flow Music, AI Studio, and Vids. It does not publish a per-track price on that page. API pricing is on Google’s Gemini API pricing page and can change. Check the live meter before you quote a client a cost.
Should I use Clip or Lyria 3.5 for Shorts?
Use Clip to find the bed. Cut the 15-second Short from a 30-second file. Promote to Lyria 3.5 only when you need a longer cue, sung verses, or a timestamped structure. A 3-minute song is the wrong first generate for a Reel.
Can I edit one section of a Lyria 3.5 song?
Not in the current API. Google documents music generation as a single-turn process. Iterative, multi-prompt editing of a generated clip is not supported. If you need section rewrite, that is a different product (for example ElevenMusic Composer), not a Lyria follow-up turn.
Can I feed product photos into Lyria?
The API allows up to 10 images with a text prompt so the model can compose from the mood and colors. That is not a guarantee the label, logo, or SKU silhouette will be “heard” in the music. Lock identity in the stills and in the lyric packet; treat images as atmosphere.
Does ClipCanva run Lyria 3.5?
No. ClipCanva is independent of Google. Use Gemini or the Gemini API for Lyria. Use ClipCanva to lock lyrics, write the video script, and generate the picture after the audio is approved.
Sources
- Create your best tracks yet with Lyria 3.5 in Gemini (Google, 4 Sep 2026)
- Generate music with Lyria 3.5 (Gemini API docs, updated 4 Sep 2026)
- Lyria (Google DeepMind)
- Introducing Lyria 3.5 in Google Flow Music (29 Jul 2026)
- Canva AI Music Generator