Making a 60-second short video,
the visuals might be edited within half an hour.
Yet the final bottleneck is often:
background music.
After sifting through free music libraries,
you may find your favorite songs are not always commercially usable.
The mood might fit,
but the length is off.
If you don’t compose music yourself,
hiring musicians isn’t practical for every short clip.
Google now brings:
Lyria 3.5
directly into Gemini.
This makes the task of "creating a perfectly fitting soundtrack for specific content"
something anyone can easily test.
What is Lyria 3.5?
Lyria is an AI music generation model from Google DeepMind.
It doesn’t analyze existing songs.
You simply describe
the kind of music you want.
For example:
"Create a 60-second upbeat instrumental that feels warm, natural, and handcrafted, featuring acoustic guitar and light percussion. The first 10 seconds are simple, gradually increasing rhythm afterward."
Lyria then truly
generates a fresh piece of music.
The latest version,
Lyria 3.5,
makes this experience even more complete.
On September 4, Google announced that Lyria 3.5 is now integrated into:
Gemini App,
Gemini API,
Google Flow Music,
and even Google Vids.
The most immediate update: Choose the style of song you want
Within Gemini,
you can now directly specify:
Genre,
like Lo-fi, Pop, Jazz, Funk, Electronic, or Acoustic.
You can also specify:
Vocal,
with singing, or
Instrumental,
purely music.
Templates are available
to quickly start from common use cases like:
background music, birthday songs, brand jingles, and more.
For non-musicians,
this is much easier than
having to describe complex music theory.
The second useful new feature: Length matched to your project
AI music often had one big issue:
You make a 45-second video,
but AI generates 30 seconds, or
a full song where only a small part suits your clip.
Lyria 3.5 now lets you control
the track length in your prompt.
During demonstrations, Google DeepMind showed that Lyria 3.5
can generate music up to about 3 minutes long.
This lets you tailor the soundtrack precisely.
Use cases include:
15–30 seconds—for brand intros,
60 seconds—for Reels or Shorts,
90 seconds—for product demos,
2–3 minutes—for full tutorials, event videos, or songs.
This is very practical for creators.
You’re no longer stuck trying to fit music to video,
but can first decide how long the video is, then ask AI for matching music.
The third interesting feature: Photos can also serve as music prompts
Lyria understands not only
text but also
photos.
Gemini Help explains you can provide
photos or videos
for context.
For example, if you upload
a photo of a sunset by the sea,
you could ask:
"Create a 70-second instrumental track inspired by this photo’s atmosphere. The first half is quiet with acoustic guitar and soft pads; the last 20 seconds brighten gradually but avoid becoming an intense commercial jingle."
This is much more specific
than just typing "make a comfortable piece of music."
Your visual materials
begin serving as
music briefs.
Especially convenient for short video creators
Imagine you’ve already edited a coffee shop Reel.
Showing:
bean grinding, brewing, steam, serving cups, and customers sitting.
Previously, the next task would be
searching for music.
Now, you can hand over
key video frames to Lyria,
and instruct:
This video is 45 seconds, no vocals, quiet in the beginning,
then rhythm builds after 20 seconds, ending cleanly.
This means
music works around the video,
not the video forced to fit pre-existing music.
Lyria can even generate complete vocals
Lyria 3.5 is more than
a background music generator.
Google states the new version improves
vocal quality, lyrics,
and musical arrangements,
and can create full songs with
verse, chorus, and bridge structures.
If you want to create
brand songs, event songs, or birthday tunes,
you can specify
male or female voice, vocal range, voice texture, instruments, tempo, dynamics, and lyric themes.
But for most small businesses, start with instrumental tracks
The reason is simple.
If you just need
product videos, tutorials, podcast intros, event recaps, or short brand clips,
what you really want is
music that enhances the mood of the visuals,
not lyrics that divert attention.
Instrumentals are usually easier
to coexist with voiceovers,
don’t compete with subtitles,
and avoid conflicting song meanings.
They’re also easier
to quickly approve for use.
So, for a first test,
don’t immediately ask AI to write a full branded theme song.
Start with a 30–60 second instrumental track.
It’s more practical.
No need for professional-level prompts, but include at least five details
Google DeepMind’s prompt guide suggests you can adjust results by
genre, tempo, instruments, dynamics, and vocals.
For general users, this can be simplified into five questions:
1. What is it for? (e.g., a 60-second handmade product video)
2. What mood? (warm, quiet, classy, lively, nostalgic)
3. Vocal or instrumental?
4. What instruments? (acoustic guitar, piano, strings, electronic drums, bass)
5. Length? (e.g., 60 seconds)
An example prompt
If you’re making
a pottery studio product video, you might say:
"Create a 60-second instrumental background track for a handmade pottery brand video. Warm, natural, quiet but not sad, featuring acoustic guitar, soft piano, and light percussion. Keep the first 10 seconds simple, add some rhythm after 20 seconds, and finish naturally in the last 5 seconds. No vocals, no dramatic parts."
This is enough to get started.
If you don’t like the first result, adjust tempo, instruments, mood, or rhythm density.
A key limitation: The API is not a conversational music editing tool
Google’s Gemini API documentation clearly states that Lyria 3.5 music generation is
single-turn,
meaning you generate once.
If you don’t like it, you generate again.
It is not a multi-turn editing tool where you can say,
"Remove the drums at 35 seconds," or
"Keep everything except soften the final chorus,"
and iteratively refine the same audio.
Don’t think of it as
a full DAW or professional music production software.
It’s more like
AI quickly providing a music draft.
Generating multiple times with the same prompt won’t yield identical tracks
Google reminds users that
results vary.
The same prompt can produce
different music.
This has pros and cons.
The advantage is you can experiment with many directions quickly.
The downside is that if you love a version, don’t assume
"I’ll generate it again later."
Save the version you like right away.
Your creations in Gemini can be downloaded immediately
According to Google Gemini Help,
once you generate music, you can
download, share, or post to social media.
Downloads are available as
MP3 audio
or MP4 with cover art.
If you just need music for videos,
MP3 is the most straightforward choice,
then import it into your video editing software.
Google Vids now supports Lyria 3.5 as well
This fits perfectly with the Google Vids series we covered yesterday on SasaDaily.
Yesterday we talked about
turning Docs, PDFs, and Word files
into summary videos.
Now with Lyria 3.5 integrated into Google Vids,
Google is gradually uniting
documents, scripts, narration, visuals, and music
into one cohesive content creation workflow.
In the past, each part required
different tools.
Now AI is linking the entire production chain.
For solo creators, the most practical use cases are simple
For example, if you make
three Reels videos per week,
each time you’d normally
search for background music,
listen, compare, check lengths,
and edit.
If Lyria can generate
three 60-second instrumentals based on your video’s theme,
all you do is pick one.
That’s a classic example of
AI giving time back to humans.
AI doesn’t decide what your brand should sound like,
it just reduces time spent repeatedly searching for music.
Second use case: fixed podcast or YouTube intros
Many small content brands don’t have
custom opening music,
thinking it’s unnecessary for non-major media.
Lyria 3.5 lets you make
opening music of 10, 15, or 30 seconds
to test for your brand.
If your brand grows later,
you can hire professionals then.
AI's perfect role is
getting you from zero to a first draft.
Third use case: events and presentations
For company year-end parties, school project videos, wedding recaps, product launches, or community events,
you might already have photos and edited videos,
but the music never quite fits.
You can provide Lyria
with your event’s main visual style,
asking it to create music
that matches your event’s vibe based on colors, atmosphere, and content.
This approach of
Image → Music
is a fresh, interesting workflow.
Fourth use case: creating jingle prototypes
If you’re a small store, e-commerce brand, YouTube channel, or podcast,
wanting a sonic identity that audiences recognize instantly,
Lyria lets you prototype jingles.
Try creating a 5-second intro,
a 15-second brand segment,
or a 30-second full version.
Test which version fits your brand best.
However, a critical note:
Just because prototypes are easy to make, it doesn’t automatically resolve trademark or music rights issues.
If using music as long-term
brand assets or commercials,
always double-check platform terms and
copyright and trademark laws in your country.
Avoid asking for "sounding like a specific singer"
Google already restricts this.
The Gemini API documentation notes that safety filters block requests for
specific artist voices or
lyrics protected by copyright.
This is a reasonable safeguard.
Instead, describe
musical characteristics.
Don’t say:
"Make it sound like so-and-so."
Say:
"Deep male voice, soulful, slow tempo, clean acoustic guitar, warm analog tones."
This approach helps you define
your own unique sound direction.
The most valuable lesson in AI music
Don’t prompt:
"Sound like who?"
Start prompts with:
"What traits do I want?"
What genre? How fast should the tempo be? What mood changes? What instruments? How are vocals performed? Which parts are quiet? Where does energy rise?
This not only keeps you safe but also
gives you control over
your creative direction.
Google adds SynthID to generated music
Google DeepMind notes that all music generated by Lyria includes
SynthID,
a digital watermark undetectable to the human ear.
Its purpose is to identify
whether content was AI-generated or edited.
So downloaded music is not
"AI-free," but contains
imperceptible SynthID embedded in the audio.
This is part of Google’s broader
provenance strategy for AI media.
Does SynthID solve all copyright issues?
No.
SynthID answers:
"Is this AI-generated content?"
It doesn’t automatically answer
whether you have unrestricted rights to use it
across countries, platforms, or commercial contexts.
Don’t confuse
AI watermark (source identification)
with commercial rights (usage permissions).
Don’t use AI-generated music blindly
Music isn’t like a piece of text
you can skim through.
Always listen to the entire track.
Especially check vocals, lyrics, and the ending
for weird sounds or sudden unwanted emotions.
If it’s for a video, check if
voiceover is clear and not overwhelmed.
And if musical climaxes occur at the right moments visually.
Simple evaluation: don’t ask if it “sounds good” first
Ask instead:
"Does this music fulfill the purpose of this content?"
For example, a 45-second product Reel might need
attention-grabbing in the first 5 seconds,
no competing voiceover in the middle,
more emotional lift after 30 seconds,
and a clean finish.
Even if another song sounds nicer on its own,
if it’s unsuitable for the video,
it’s not the better choice.
AI tools must always serve
your work goals.
For a first try, focus on a 60-second instrumental only
Don’t immediately dive into
full songs, lyrics, vocals, jingles, podcasts, or ads all at once.
Pick a short existing video,
note its length,
give Lyria details on purpose, mood, instruments, length,
request no vocals,
generate several versions,
then test in your video.
You’ll quickly learn if it truly saves time on soundtrack search.
A very simple initial test
For example:
60-second video
Theme: handmade products
Prompt:
Warm, natural, instrumental, acoustic guitar + piano + light percussion, slow beginning, add rhythm mid-section, natural ending, no vocals.
Generate,
download MP3,
drop into the video,
watch the whole thing,
then ask:
"Does this music make the video feel complete?"
If yes, keep it.
If no, try another direction.
Don’t force it just because it’s AI-produced.
The biggest difference between Lyria 3.5 and typical "AI music recommendation"
Music recommendation searches
existing libraries.
Lyria generates
new music matching your task.
This signals a major shift in content creation:
In the past,
you had content first, then searched for music.
Now,
you have a need, and AI creates
a custom first draft.
This is happening with images, videos,
and now music.
But professional musicians remain invaluable
On the contrary, AI excels at
quickly generating ideas, prototypes, background materials, and cost-effective experiments.
For brand audio identity,
complete arrangements, precise mixing,
big advertising campaigns, long-term music identity,
emotional vocals, and storytelling,
professional musicians provide
selection, aesthetics, expression, and final quality.
Just like AI images,
good visual direction remains essential.
The real value of Lyria 3.5 today isn’t "AI finally writes songs"
AI has been capable of music generation for a long time.
The real breakthrough is
its integration
into workflows accessible to everyday users,
like Gemini, Google Vids, Flow Music, AI Studio, and APIs.
Meaning,
music generation is moving from a novelty demo
into a standard component of content creation workflows.
Need music for a video?
Generate it instantly.
Want music to match photos' mood?
Generate it directly.
Need a specific length background track?
Generate it promptly.
This is the truly practical aspect today.
Finally, remember this usage principle
Don’t ask first:
"How amazing is Lyria at writing music?"
Ask instead:
"What parts of my workflow keep wasting time looking for music?"
Could be
Reels, YouTube, Podcasts, event videos, product videos, or presentations.
When you have real work,
Lyria 3.5 can transform from
a fun AI toy
into
a genuine time-saving tool.
Today, grow a little with AI.
Learn a new AI skill every day.
Save a little time each day.
Improve your abilities bit by bit.
SasaDaily grows with you.
Recommended reading
AI Can Imitate a Style in Seconds, but Creators May Have Spent Ten Years to Get There
