You have the episode recorded, the levels balanced, and the edit locked. Then you hit the part nobody warns new podcasters about: the music. You need an intro that sounds like your show, an outro that plays under the credits, and a short bed for the ad read or the segment change — and you need all of it to be legal to publish, on every platform, forever. The free loop you found last month turned out to require attribution you can't fit into audio. Hiring a composer for a fifteen-second sting feels absurd. And the "royalty free podcast music" pack you almost bought had a license clause that made your stomach drop.
This guide fixes that. Instead of digging through libraries and decoding license tiers, you generate the exact cues your show needs — an intro, an outro, and loopable beds — built around your podcast's mood and length. The examples use Google's Lyria 3 Pro because it does the two things podcast audio depends on: it holds real musical structure so an intro builds instead of just starting, and it lets you place section changes at exact timestamps so a cue lands where your voice comes in. Where platform terms or pricing matter, I keep it qualitative and tell you what to verify, because those move faster than any blog post.
Why podcasters should generate cues, not hunt for them
A stock library hands you a finished track and asks your show to bend around it. Generation flips that. You describe the role — "twelve-second intro that resolves right before the host speaks" — and get audio shaped to that job. For a podcast, where the same three or four cues repeat every single episode and become part of your brand, that fit compounds.
| What breaks podcast audio | Stock library | AI-generated cue |
|---|---|---|
| Wrong length | Trim or awkward fade | Request the exact seconds you need |
| Generic identity | Same loop as other shows | Every generation is unique to your prompt |
| Attribution strings | Credit lines, tracking IDs | One platform's terms you can read up front |
| Bed fights the voice | Fixed mix, vocals intrude | End the prompt with Instrumental. for a clean bed |
The trade-off is honest. A generator won't replace a composer scoring a narrative documentary, and today Google's model does not export separated stems — so if you finish in a DAW and need to solo the drums under an ad read, that's a real limit. But for intros, outros, transition stings, and ambient beds — the bulk of what a podcast needs — a purpose-built cue beats a borrowed loop on fit and peace of mind.
Step 1: Match your show's mood before anything else
The fastest route to a bad intro is prompting for a genre and stopping there. "Upbeat music" gives you a texture, not an identity. Google publishes a prompt formula worth memorizing:
[Genre & style] + [Mood] + [Instrumentation] + [Tempo & rhythm] + [Vocal style & language] + [Lyrics]
Most podcast cues are instrumental so they never compete with speech, which means you almost always end the prompt with Instrumental.. A true-crime show and a startup interview show want opposite feelings, and the prompt is where you set that:
Tense cinematic intro, dark and suspenseful, low pulsing
synth bass with sparse piano and a distant string swell,
slow around 70 BPM, building quiet dread. Instrumental.Bright modern intro for a business interview show, confident
and optimistic, clean electric piano with light percussion
and a warm bass groove, around 110 BPM, energetic but not
busy. Instrumental.Notice how much of each prompt is about feeling and role — "building quiet dread," "energetic but not busy" — rather than just style. That is the difference between a cue that sounds like your show and one that sounds like everyone's show.
If you would rather not stare at a blank prompt box, our step-by-step walkthrough for turning a prompt into a finished track covers the full loop, and the prompt guide breaks each slot of the formula down with copy-paste examples.
Want to hear it on your own show first? You can open the generator and describe your podcast's vibe in one line before you read further — the rest of this guide makes more sense once you have one intro in your ears.
Step 2: Generate the four cues every show reuses
Think of your podcast's music as a small kit, not one track. Four cues cover nearly every show, and each is a separate quick generation:
- The intro — 10 to 20 seconds that establishes your identity and resolves cleanly right before the host starts talking.
- The outro — a warmer or more conclusive version of the intro theme to play under credits and your call to subscribe.
- The transition sting — 2 to 4 seconds to mark a segment change or the jump into an ad read.
- The bed — a low, unobtrusive loopable ambient track to sit under a sponsor read or a reflective monologue.
Lyria 3 Pro generates original pieces up to three minutes in a single request, with real intros, verses, and resolutions rather than a repeated loop — far more room than any cue needs, which means your intro can genuinely build and land instead of just fading in. For the short cues, the base Lyria 3 model produces up to 30-second pieces, which is perfect for prototyping a sting or a bumper before you commit a full Pro generation to the keeper. Prototype on the short model, produce on Pro.
For a loopable bed, describe the loop in the prompt — ask for a steady, evolving-but-repeating texture with no dramatic peak — so it sits under speech without pulling attention. Generate a couple and trust your ear.
Step 3: Time the intro so the music hands off to the voice
This is the feature that separates a real intro from a clip that just stops, and it is the one most generators cannot do. Lyria 3 Pro accepts timestamp cues, so you script where the arrangement changes instead of hoping the swell lands somewhere useful:
Warm indie-folk intro for a storytelling podcast.
[00:00] single fingerpicked acoustic guitar, gentle and inviting
[00:06] add soft brushed drums and a light bass
[00:12] a short uplifting swell, then settle to a quiet sustained
note ready for a voice to enter. Instrumental.Line that final timestamp up with the exact moment your host speaks, and the music appears to introduce them rather than collide with them. Google's official prompting guide documents this timestamp technique with a worked example; for podcast intros it is the single highest-leverage move, and almost nobody uses it. Keep your cue points a few seconds apart so transitions breathe, and set the last timestamp to your target length so the intro resolves instead of cutting off.
Step 4: When Lyria RealTime fits instead
If your show is live, or you want music that reacts to a conversation in the moment rather than a fixed pre-rendered cue, there is a different tool for that. Lyria RealTime is Google's interactive model: it streams continuous 48 kHz stereo audio with low latency — roughly two seconds at most between a control change and the effect you hear — and lets you steer key, tempo, density, and brightness on the fly by blending prompts. It is instrumental only, no vocals, which suits a live backing bed perfectly.
For a normal edited podcast that is overkill — you want a repeatable, finished cue, so Lyria 3 Pro wins. But for a live-streamed show or an interactive segment, RealTime is the model that fits. Our overview of what Lyria RealTime does covers where it makes sense.
Rule of thumb: if the cue is the same every episode, generate it once with Lyria 3 Pro and reuse it. If the music needs to change while you are recording, reach for Lyria RealTime.
Step 5: Understand SynthID and the rights question before you publish
Here is where podcasters get nervous, and where you should be careful rather than casual — especially on the phrase "royalty free podcast music," which people treat as a guarantee when it is really just a licensing term.
Every track Google's Lyria models generate carries an invisible SynthID watermark identifying it as AI-generated, and outputs support C2PA content credentials. SynthID is a provenance signal baked into the audio — it does not stop you from using the track, and it is not a copyright strike waiting to happen. It exists so the origin of the audio can be identified, which is increasingly what platforms and listeners expect from AI content.
On rights: Google states its models are trained on licensed and permissible data, and generated outputs are checked against existing content. That is a meaningfully stronger starting point than a mystery-sourced "free music" download. But whether you can use a specific track in a monetized or sponsored show depends on the terms of the access path you generated it through, and those terms evolve. Before you attach a generated cue to a published episode:
- Read the terms for your access path. Rights differ between the Gemini app, the API, and credit-based studios. Confirm what commercial use your plan grants.
- Check your host and ad network's policies. If you run dynamic ads or distribute through a network, verify their current stance on AI-generated audio.
- Keep your generation records. The prompt, the date, and the C2PA credentials are your paper trail if a claim ever comes up.
- Don't assume "royalty free" from one source transfers. The phrase means nothing without the actual terms behind it; each tool's license is its own.
None of this is legal advice — it is a checklist. When money or a sponsor is involved, verify the current terms yourself. Our deeper guide to commercial use of AI music walks through the same ground with more room.
A low-friction test before you re-score your show
Don't rebuild your whole audio kit on faith. Run one quick check first: generate a single 15-second intro aimed at your show's mood, drop it in front of your actual cold open, and listen once. Does it feel like your show? Does it hand off cleanly to the first line? If yes, generate the matching outro, sting, and bed — a full kit in an afternoon. If no, you've spent thirty seconds learning that instead of re-editing a finished episode.
Frequently asked questions
Is AI-generated music royalty free for podcasts? Not automatically — "royalty free" is a licensing term, not a property of AI audio. What you can do with a track is set by the terms of the access path you generated it through. Read those terms and confirm your host and ad network's AI policies before you publish a monetized episode.
How long can a podcast intro or bed be? Lyria 3 Pro generates up to three minutes in one request with full musical structure — far more than any cue needs, so an intro can genuinely build and resolve. The base Lyria 3 model does up to 30 seconds, which is ideal for stings, bumpers, and testing.
Can I make instrumental-only beds so they don't clash with speech?
Yes. End your prompt with Instrumental. and the model omits vocals, giving you a clean bed to talk over during ad reads, intros, and reflective segments.
Will the SynthID watermark cause problems on podcast platforms? SynthID is a provenance marker, not a copyright flag. It identifies the audio as AI-generated but does not by itself block use or trigger a claim. Whether you can monetize depends on your license and platform policy, not the watermark.
Do I need a Google Cloud account to use it? No. Beyond Google's own surfaces, credit-based studios like this one run Lyria 3 Pro on pay-as-you-go credits with no cloud setup — see pricing for current credit costs.
Build your show's sound today
Custom podcast music used to mean a composer or a compromise. Now it means a handful of good prompts, a few timestamps aligned to where your voice enters, and a short listen before you publish. Match your mood, generate the four cues your show reuses, time the intro's hand-off, and check the terms for your access path — that is the whole workflow.
When you are ready to hear it on your own show, head to the generator, describe your podcast's vibe, and get a fitted intro in the length you need. If you want the deeper background on the model first, the Lyria 3 Pro guide covers what it can and can't do.
