The format is short and vertical, but what makes it a drama is that the same people, places, and unresolved questions come back next episode. That is the part generation tools usually drop, and the part Tosheo is built around.
What an AI microdrama actually is
A microdrama is a serialized drama told in short, vertical episodes. Each episode opens on a hook, moves through a small number of scenes, and ends on a question the next episode has to answer. Audiences watch them the way they read chapters, in a sequence, on a phone.
The “AI” part describes how the images, voices, and shots are produced. It does not describe who decides what happens. In a microdrama the story still has to work: a character has to want something, someone has to stand in the way, and the ending of episode three has to make episode four feel necessary.
The vocabulary around this format is still settling. These terms are used more or less interchangeably in the market, and this site treats them as the same thing:
Why it is not a short film, and not a clip
This distinction decides almost every production choice that follows, so it is worth being precise about. A short film and a microdrama episode can be the same length and still be different objects.
- Resolves. The audience leaves satisfied and does not need more.
- Introduces its characters from scratch and can afford to.
- Judged as one finished artefact.
- Production risk is concentrated in a single piece of work.
- Continuity only has to hold for its own runtime.
- Deliberately withholds. Its job is to earn the next view.
- Assumes the audience already knows the cast.
- Judged on whether it belongs to the episodes around it.
- Risk is spread across many small, correctable units.
- Continuity has to hold across everything already released.
A clip sits outside both. It is complete on its own and belongs to nothing, which is why a beautiful clip is not evidence that a season is achievable.
Why a season is the unit of production
Generating one striking sixty second video is a solved problem. Generating the twentieth episode of a season, in which the lead wears the same jacket, speaks in the same voice, stands in the same kitchen, and remembers what happened in episode four, is not.
The reason is arithmetic. Every episode you add multiplies the number of things that must stay true. One episode has a handful of facts to honour. By episode ten there are the accumulated facts of nine previous episodes, and each of them is an opportunity to contradict something the audience remembers better than you do.
A production that treats each episode as a fresh prompt accumulates those contradictions until the audience stops believing the world. Continuity is not a finishing touch on a microdrama. It is the product.
The seven levels a season is built from
Tosheo models a production as seven nested levels. The point of naming them is that a decision made at one level should not have to be re-made at every level below it, and a mistake at one level should be fixable without touching the others.
- 01 · World The place, its rules, and its visual language.
- 02 · Series The premise and the engine that generates stories inside that world.
- 03 · Season The arc: what changes, in what order, and what it costs.
- 04 · Episode One hook, a small number of scenes, one unresolved ending.
- 05 · Sequence A run of scenes that completes one movement of the episode.
- 06 · Scene One location, one continuous stretch of time.
- 07 · Shot The smallest unit that can be reviewed, rejected, and replaced.
Most of the pain in generated video comes from working only at the bottom of that list. If the only unit you can address is the shot, then the world, the cast, and the arc are being re-described every time you generate, and they will drift.
What has to stay consistent, specifically
“Continuity” is vague enough to be useless as a promise, so here is the concrete list of properties Tosheo tracks across a season:
- Face and body. Approved references travel with the character into each shot, and generated results are evaluated against them.
- Expressions. The specific emotional register a character has been approved to play, so a stoic lead does not suddenly mug.
- Costume state. What they are wearing now, including damage, weather, and changes that happened on screen.
- Voice and pronunciation. The approved voice, local phrasing, names, and pronunciation guidance. A new voice requires creator review.
- Locations and sets. Reusable places with fixed geometry, so a room does not rearrange itself between visits.
- Significant props. Objects the plot depends on, tracked as deliberately as characters are.
- Relationships. How two characters stand with each other at this point in the season.
- Story state. What happened, who witnessed it, what changed, and what is still unresolved.
Tosheo actively tracks and evaluates these properties and surfaces contradictions for review before generating, but it does not promise perfection. You retain final approval over every episode, which is the only guarantee worth offering.
What that looks like in practice
The demonstration season on this site is a useful illustration because it is deliberately small: two recurring characters, three locations, and one object that carries the plot.
A woman returns home with a brass keepsake. Inside it is a train ticket dated tomorrow, in a name that is not hers.
- Recurring cast. Anaya, the returning daughter. Kabir, the friend who stayed. Both approved once, with faces, costume states, and voices.
- Locations. The railway platform, the family home, and the street between them. Reused every episode rather than re-imagined.
- The prop that matters. The brass keepsake. It is tracked, because a later episode depends on exactly what was inside it and who saw it.
- What episode two inherits. The cast, the three locations, the keepsake, the ticket, and the fact that Anaya has seen the name on it and Kabir has not.
That last row is the whole argument for treating a season as one production. None of it is re-described in episode two. It is carried.
Regional stories and language
Tosheo is built for regional storytellers, and language is treated as part of continuity rather than a translation step applied at the end. Voices, local phrasing, names, and pronunciation guidance are approved once and carried through the season.
This matters more in regional production than it does in English. A name mispronounced inconsistently, or a register that slides between formal and familiar without reason, reads as carelessness to an audience that speaks the language and is invisible to one that does not.
The pilot begins with one focused launch language during validation. Additional regional languages will be considered after production quality and economics are proven, and this site does not list languages it has not proven.
How a season actually gets made
Tosheo separates the decisions from the rendering. First you establish and approve the things that must not drift: the premise and world rules, the recurring cast, the language and voices, and the shape of the season. Only then does an episode get written, storyboarded, checked, costed, and rendered.
You approve each major stage before it becomes canon, and you see a production estimate before anything is rendered. When a single shot comes back wrong, the goal is to replace that shot rather than regenerate the episode around it, so accepted work survives.
Tosheo is in closed validation and is recruiting a focused founding creator cohort. Applications are open.
Questions creators ask
What is an AI microdrama?
A serialized drama told in short vertical episodes, where the images, voices, and shots are produced with AI while the story, cast, and approvals stay with the creator. Its defining feature is that the same characters and world return in every episode.
How is a microdrama different from a short film?
A short film resolves and stands alone. A microdrama episode deliberately withholds, assumes the audience already knows the cast, and is judged on whether it belongs with the episodes around it. Its production problem is consistency across many episodes rather than one finished piece.
Is microdrama the same as micro drama, mini drama, or short drama?
In practice yes. The vocabulary is still settling and these terms are used interchangeably for short, vertical, serialized drama. This site treats them as the same format.
Do I need a finished script to start?
No. You can begin with a one-line idea, a character, or a conflict, and Tosheo can propose a reviewable series foundation for you to accept or reject. If you already have a synopsis, screenplay, novel, comic, or series bible, you can start from that instead.
Can the same characters carry a full season?
That is the core design goal. You approve a recurring cast once, including faces, costume states, voices, and relationships, and those approved references carry into later episodes rather than being re-imagined each time.
Is continuity guaranteed?
No. Continuity is actively managed and evaluated, and contradictions are surfaced for review before generation, but it is not promised as perfect. You retain final approval over every episode.
Which languages does Tosheo support?
The pilot begins with one focused launch language during validation. Additional regional languages will be considered after production quality and economics are proven.