Write the script, generate a clean voiceover, then generate a music bed that supports it instead of competing with it. Picasso IA chains text to speech and music generation from one account, so a 30-second spot comes together without booking a studio or a composer.
A thirty-second radio spot is a script and a soundtrack fighting for the same handful of seconds of attention, and getting both right used to mean booking a studio, a voice actor and a composer for something that might run for two weeks. Picasso IA generates a clean voiceover and a matching music bed from the same account, then hands both files to any audio editor for the final mix.
A radio or podcast ad almost always works in two layers, and treating them separately is what makes the spot sound professional instead of homemade. The first layer is the voiceover, which does the actual selling: it states who you are, what you are offering, and what the listener should do next. The second layer is the music bed underneath it, which sets the mood and keeps the ad from sounding like a dead-air announcement, but it is not supposed to be noticed on its own terms.
That second-layer discipline is where a lot of amateur spots go wrong. A music bed with a strong vocal hook, a busy drum fill or a loud chorus right where the voice delivers the price competes with the read instead of supporting it. The bed's job is to occupy the space around the voice, not the space the voice needs.
On Picasso IA the two layers come from two different capabilities in the same Toolkit: a text-to-speech model turns your script into a voiceover, and an AI music generation model turns a short style prompt into an instrumental bed. Both are generated separately and combined afterward, which is also what gives you control over the balance between them.
Radio and podcast ads are sold in fixed lengths, usually 15, 30 or 60 seconds, and a script that runs long gets cut off mid-sentence or read at an unnatural sprint. Average conversational read speed is close to 150 words per minute, roughly two and a half words per second, so a 30-second slot comfortably holds about 65 to 75 words once you leave a little room at the start and end.
Write to that word count rather than writing everything you want to say and hoping it fits. A script that only just fits gets rushed in the read, and a rushed voiceover sounds anxious rather than confident, which works against you when the whole point of the ad is to sound trustworthy. Leave a beat of silence after the offer and before the call to action, since a pause is often what makes the final line land.
Generate the voiceover before you finalize the script, not after. Listening to the actual timed read tells you immediately whether the pacing works, in a way that reading the script silently to yourself never quite does.
The tone of the script should decide the tone of the music, not the other way around. A price-driven retail promotion and a calm professional service message are both valid ad tones, but a bed that suits one will undercut the other if you swap them.
| Ad tone | What the script sounds like | Matching music bed |
|---|---|---|
| Upbeat retail or sale | fast pace, exclamation, a clear deadline | bright tempo, major key, light percussion |
| Calm professional service | measured pace, reassuring, few adjectives | soft pads, slow tempo, minimal percussion |
| Urgent limited-time offer | short sentences, repeated deadline, direct | driving rhythm, rising tension, no vocals |
| Warm local or community business | conversational, specific, a little informal | acoustic instruments, gentle tempo, familiar tone |
Run a few candidate beds against the same voiceover before picking one. A bed that sounds great alone can still fight the read once the two are layered, and the only reliable way to catch that is to listen to them together.
Pro tip: ask specifically for an instrumental bed with no vocals and describe the arrangement as sparse in the same frequency range as a human voice, roughly 300 Hz to 3 kHz. A bed built from bass, wide pads and light percussion leaves that middle band open for the read, so the voiceover stays intelligible without you having to fight the mix later.
Intelligibility is the one thing a radio ad cannot compromise on, because a listener who cannot make out the offer will not remember the brand either. That means the voiceover has to sit clearly above the music bed for the entire length of the spot, not just at the loudest moments.
Generation gives you a head start on this, but the final balance is a mixing decision made in an audio editor, not something either model does for you. A typical approach lowers the music bed several decibels under the voice track, often called ducking, and pulls it down further during the busiest phrases of the read. Picasso IA produces the two source files at full quality; the ducking, EQ and loudness leveling that make them sit together happen afterward, in whatever audio tool you already use.
Choosing a sparse bed to begin with makes that mixing pass much easier, which is why the instrumental-and-sparse-arrangement advice above matters as much during generation as it does during the mix itself.
Generating the pieces takes minutes once the script is settled; the script itself is usually where the real time goes; and every step below can be redone individually if one part is not landing.
An honest tool page tells you where the tool's job ends, and five limits come up consistently with generated audio commercials.
None of these limits change what the tool is for, generating the two audio layers of a commercial quickly and cheaply so the creative decisions happen on the page instead of in a booked studio; they mark where its job ends and a mixing pass begins. Beyond radio spots, the same text-to-speech and music generation models cover audiobook narration and background music for longer videos, and the model catalog lists every voice and music model available.
It generates the two layers separately rather than as one combined file, which is actually the better workflow. A text-to-speech model produces the voiceover from your script, and a music generation model produces an instrumental bed from a style description. You then combine them in an audio editor, which gives you control over the balance instead of accepting whatever ratio a single combined output happened to produce.
Aim for roughly 65 to 75 words, based on an average conversational pace of about 150 words per minute. That leaves a small margin at the start and end so the read does not feel rushed. Generate the voiceover early and check its actual timed length, since punctuation and phrasing shift the pace slightly from one script to another even at the same word count.
Modern text-to-speech voices sound close to a human read for straightforward, well-punctuated scripts, and the gap is smallest on short, clear commercial copy like a radio spot. It is still worth listening critically to unusual words, brand names and numbers, which are the places a synthetic voice is most likely to slip. Testing two or three voices on your exact script, rather than picking blind from a list, is the fastest way to catch that.
Picasso IA's voice catalog includes multiple accents and languages, so you can often match a market without extra work. Listen to a sample of the shortlisted voices reading a line close to your actual script, because accent quality varies by voice and by the specific words involved, not just by the accent label attached to it.
You control that in the prompt. For a radio ad the almost-always-correct choice is instrumental, since a music bed with its own vocal line competes directly with the voiceover for the listener's attention. Ask explicitly for an instrumental arrangement when you generate the bed, and describe the mood and tempo rather than requesting a song.
Start with a sparse bed, generated with light percussion and space in the middle frequency range where speech sits, and then duck the music under the voice in your audio editor. Ducking means automatically lowering the music's volume for the duration of the voiceover and letting it come back up in the gaps. Picasso IA generates the two source files; the ducking itself happens in whatever audio editor you use to combine them.
Yes. Translate the script, adjusting the word count for the target language's typical pace rather than translating word for word, and generate a new voiceover in that language using a matching voice from the catalog. The same music bed usually works across languages unmodified, since an instrumental track has no language of its own.
There is no fixed cap; each generation uses credits, and new accounts start with free credits that cover several rounds of testing a script and comparing voices or beds. Exact costs by model are listed on the pricing page, and testing two or three voiceover takes against two or three beds before settling on a final pair is a reasonable amount of experimentation for a spot you intend to run repeatedly.
Yes, and podcast sponsorship spots are one of the most common uses of this exact workflow, since they follow the same short-script-plus-bed structure as traditional radio. The main difference is delivery format rather than production: a podcast ad is usually exported as a standalone audio file the host inserts, while a broadcast spot may need to match a station's specific loudness and format requirements.
Trim it. Precise slot timing is a mixing-stage adjustment, not a generation-stage one: cut a beat of silence from the start or end of the file, or shorten a pause between the offer and the call to action, until the final export matches the length your station or host requires. Re-generating the whole voiceover for a one- or two-second difference is rarely necessary.
A radio spot lives or dies on a clear offer and a mix that does not bury it, both cheap to test before a final cut. Start the read and the bed in the Toolkit, or see how the same models handle a longer format in AI voiceover for videos.
