September 24, 2026
How to create an audio drama series with AI: from idea to mixed episode (2026 guide)
Bible, writing with memory, voices, sound effects, ambiences, music and mixing: the complete method to produce an audio drama with AI, costed on a real pilot episode.
Frank HoubreFounder of Imaginode
Why everyone is getting into audio series
Audio series cut into short episodes that end on a cliffhanger have become a mass format, and AI has just made them something one person can produce.
People listen to them on the commute, while cooking, while running: fiction cut into episodes of a few minutes, performed by several voices, with sound effects and music that set the mood. Apps like Pocket FM claim more than 250 million listeners for this format alone, and the audience goes far beyond podcast fans.
Until now, producing a series took a team: a writer for the season, voice actors for every character, a studio to record, a sound engineer for effects, music and mixing. Weeks of work and a budget that put most writers off.
In 2026, every link in that chain has an AI equivalent: language models that write and remember a whole season, synthetic voices that can perform dialogue, sound effect and music generators. This guide shows how to put them together, step by step, with the real numbers of a pilot episode produced in Imaginode's Drama Studio.
The five building blocks of an audio series
An audio series rests on five blocks: a bible, scripts, voices, sound and a mix. Skip one and the ear notices right away.
The bible is the series reference document: who the characters are, what they want, how they talk, where the story takes place, which threads it opens and how it closes them. Without a bible, episode 4 contradicts episode 2.
The scripts are the episodes themselves, written for the ear: lines, a few narrator interventions, sound effect and music cues. The voices perform those lines, one voice per character, recognizable from one episode to the next.
Sound comes in three layers: one-off effects (a door, a phone), the continuous ambience of each location (a station, a forest at night) and music. Finally the mix places everything on a timeline, ducks the music under the voices and sets the loudness so an episode plays at the same volume as any podcast.
Step 1: turn an idea into a series bible
Two sentences are enough to get a complete bible, as long as you set the genre, the episode length and the season size.
Take the idea used for our pilot: "A conductor on the last Paris to Nice night train finds out that a passenger travels every week with a ticket dated 1987. She decides to follow him." Genre thriller, 3 minute episodes, a season of 6 episodes.
From that, the Drama Studio's Bible agent produced in 45 seconds a title ("Le Train de Minuit"), a logline, six characters including a narrator, five locations with their soundscape described for a sound engine, six story threads, a music theme and the plan of the six episodes, each ending on a twist. Measured cost: 6 credits.
The most important part of this step is casting. Each character gets their own voice, chosen by gender, age and timbre, and that voice follows them for the whole season. Read the character descriptions carefully: they also say how each one talks, and that is what will make the dialogue feel alive.
Step 2: write each episode with the memory of the season
The real challenge of a series is not writing one good episode, it is writing the seventh without forgetting what happened in the first six.
A language model simply asked to "write episode 5" makes things up. To hold a season together, the AI writer needs the bible, the season plan and a memory: the summary of every episode already written, what each character has learned, and the state of every thread. That is exactly what the Drama Studio does: after each episode it records its memory, and the next episode starts from there.
In the pilot, the 3 minute episode 1 was written in 40 seconds: 60 lines, 3 scene changes, 4 sound effects, 2 music cues, about 3,000 spoken characters. Measured cost: 7 credits. The memory noted, for example, that the conductor had scanned the 1987 ticket and that the system accepted it, a fact episode 2 could no longer ignore.
Then comes the review. A second agent, the script editor, scores the hook, pacing, coherence, dialogue and ending out of 10, and points at the faulty lines. On the pilot it caught two real continuity errors, including two conductors talking to each other from two different carriages without a phone. Cost: 2 to 3 credits. Its notes feed a rewrite in one click, and the previous version can be restored. For the writing rules themselves, see our article on how to write an audio drama that hooks.
Step 3: have the voices perform
A scene is performed in one go: all its lines go to the voice engine together, which chains the characters like a real conversation.
That is the difference between a reading and a drama. If each line is generated separately and stitched together, you hear regular gaps and intonations that do not answer each other. The ElevenLabs dialogue engine used by the studio receives up to ten lines of a scene with each character's voice, and returns a single passage where reactions flow naturally.
Each line can carry a delivery: whisper, laugh, sigh, curious, sad or excited. These tags really change the performance, and the script editor flags the ones that go against the scene, like a laugh during a tragic reveal.
Voices perform in fourteen languages, and a series can be written and produced in your audience's language. For a single voiceover or a two voice podcast, the canvas tools AI voiceover and two voice podcast are enough; for a fiction with several characters, the studio is what keeps the cast consistent from one episode to the next.
Step 4: sound effects, ambiences and music
Sound tells as much as the voices, as long as you use little of it: one ambience per location, an effect only when it carries the story, one music theme for the whole series.
Ambience is the most cost effective layer: a twenty second loop per location (a station platform at night, the inside of a moving train) is enough to place the listener, and it plays under every scene set there. In the pilot, two ambiences cover the whole episode.
Sound effects need restraint. Without a number in its instructions, the AI placed eleven of them in two minutes, and each one cut into the dialogue. The studio now caps it at about one effect per forty seconds, kept for sounds that move the story: a door slamming, a ticket handed over, a departure whistle. For standalone sounds, the canvas sound effect generator also works on its own.
The music theme is composed once, as an instrumental, and reused in every episode: it opens the episode, steps back under the voices, and a short dramatic sting underlines the ending. It is the biggest cost of the first episode (84 credits on the pilot), and the only one that does not repeat.
Step 5: mix and publish
The mix places every sound on the timeline, ducks the music under the voices and normalizes loudness around -16 LUFS, the podcast standard.
In the Drama Studio, mixing happens in your browser, free of charge: dialogue blocks follow each other at their real length, effects sit on top without holding the next line back for more than a second and a half, each ambience loops under its scene, and the music drops automatically whenever a character speaks.
The mixed pilot lasts 4 minutes, measures -16.2 LUFS with a true peak at -1.7 dBFS, and weighs under 4 MB as an .m4a. While you listen in the studio, the current line lights up in the script, so you can spot the sentence to rework in a second.
The file is yours: download it and publish it on a podcast platform, on YouTube with a still image, or on your website. It is also stored in your Imaginode media, where it can become the soundtrack of a video edited on the canvas.
How much an episode really costs
On the 3 minute pilot, writing cost 15 credits and production 151, including 84 for a music theme that is only paid once.
The breakdown measured on 23 September 2026: season bible 6 credits, episode writing 7 credits, review 2 to 3 credits. In production, 13 dialogue passages, 4 sound effects, 2 ambiences, one dramatic sting and the music theme, that is 21 sounds and 151 credits. The mix cost nothing.
From the second episode on, the theme and the ambiences of locations already visited are reused: a 3 minute episode then comes to about 60 to 70 credits of sound, plus about 10 credits of writing. For voices alone, count about 12 credits per minute of dialogue.
And a sound already produced is never paid for twice: if you fix a line, only the passage containing it is regenerated. That is what keeps rewrites affordable, even late in production.
The mistakes that make listeners drop off
Too much narrator, characters who sound alike, sound effects everywhere and an episode with no cliffhanger: four mistakes to fix before production.
A narrator who tells everything turns the fiction into an audiobook: keep them for jumps in time and place, and let the characters carry the action. Two characters with voices that are too close get confused: in the bible, give them different ages, timbres or accents.
Too many sound effects tire the ear and break the rhythm: one well chosen ambience replaces ten of them. And an episode that ends without an open question gives no reason to play the next one: the script editor's "ending" score is the one to watch first.
The easiest way to start is to listen to one episode produced end to end. The Train de Minuit episode plays on the Drama Studio page, and if you are weighing several tools, our comparison with Pocket FM's Sherpa details what each one actually does.