AI Short Drama Generator

A short drama, also called a micro-drama or vertical drama, is a serialized story made for phones: vertical 9:16 video, episodes of about one to two minutes, and a cliffhanger at the end of nearly every one. A series can run from 20 to 100 episodes, about as long in total as one or two feature films. The format started in China and spread worldwide through dedicated apps.

What makes it hard to produce with AI is the cast. The same three or four people have to appear in every episode, look the same and sound the same. On Mage you save each lead once as a character, from one portrait, with a voice of up to 10 seconds. From then on, every episode casts them by @handle.

Each episode is a Story Studio run: an episode prompt of up to 1,000 characters, up to three leads, and 5 to 20 scenes filmed in order in 9:16. Scene Builder handles two-character confrontations, Lipsync puts an exact line in a character’s mouth, and Storyboard joins up to six clips into one vertical MP4.

Micro-drama episodes usually run one to two minutes, so build each one from six scenes of 10 to 15 seconds. That is 60 to 90 seconds of story, and six is also the most clips Storyboard joins into one file, so a whole episode exports as a single vertical MP4. On Cherry, pick 10 or 15 seconds per scene; Cherry 2 Pro and Lemon go up to 30 seconds when a scene needs a longer speech.

Give the six scenes fixed jobs. Scene 1 is the hook: open on the most dramatic image of the episode, because viewers decide in the first seconds. Scene 2 reminds them who wants what. Scenes 3 and 4 escalate: a lie, a confrontation, a secret overheard. Scene 5 is the reversal. Scene 6 is the cliffhanger: cut on the question, never on the answer.

Story Studio writes the scenes so the story prompt is satisfied by the last one, so tell it where this episode stops. End the episode prompt with the cliffhanger itself, such as "It ends as she opens the envelope and freezes", then read the last scene’s script and cut anything that answers the question too early.

• Scene 1: the hook, the episode’s strongest image.
• Scene 2: who wants what.
• Scenes 3 and 4: escalation.
• Scene 5: the reversal.
• Scene 6: the cliffhanger, cut before the answer.

Every Story Studio run starts fresh: it has not seen your last episode. Continuity comes from three things you control. The cast: the same saved characters, by @handle, in every run. The world: the penthouse, the office, the car and the signature outfits saved as references, so they look the same in episode one and episode thirty. And the recap: one line at the top of every episode prompt that says where the story stands.

A run takes up to three leads, and cast and references share five slots, so plan each episode around the three people who matter most in it. Keep a series bible outside Mage with each lead’s @handle, what they want, what they hide, and the places and outfits tied to them, and paste the relevant lines into each episode prompt.

Voices are part of the cast. A character’s saved voice keeps up to 10 seconds of audio, and Cherry, Raspberry and Lemon in Story Studio use it whenever that character speaks. If a lead has no voice yet, upload a clip or generate one from their portrait when you create or edit the character, and keep it for the whole season.

• Open every episode prompt with a one-line recap.
• Save recurring places and outfits as references.
• Keep a series bible with each @handle.
• Never change a lead’s voice mid-season.

Short dramas live on two-person scenes: the proposal, the betrayal, the confrontation in the rain. Scene Builder is built for them. Add one image of each character, pick a preset such as Dinner Date, Wedding, Late Night Drive or Mafia Meeting, or press Custom and write a one-line idea, and it writes a full prompt you can edit. It runs on Cherry by default, or on Cherry Pro, Cherry 2 Pro, Raspberry or Lemon, from 2 to 30 seconds depending on the model.

When a line must land word for word, such as a vow or a reveal, make the audio first: AI text to speech with an audio tag like [whispers], or the lead’s own cloned voice in Seed Audio, covered on the AI voice cloning page. Then open Lipsync, add the character and up to 15 seconds of audio, and it syncs the lips to the line.

Mage does the production. It writes a shot-by-shot script for every scene from your episode prompt, sizes the shots and dialogue to the clip length, films the scenes in order with your saved faces and voices, stages two-character scenes, syncs lips to recorded audio, and joins up to six clips into one file.

You do the showrunning: the premise and the series bible, the episode outlines and every cliffhanger, reading and fixing each script before you pay to film it, choosing the takes and rerunning the scenes that miss, and the finish in your editor, with captions, title cards, music and the cover. Mage does not add captions or subtitles, and publishing, platform deals and monetization are up to you.

Plan the pace of production too. You can run one Story Studio or Autocreate campaign at a time, so episodes film one after another, and Script History keeps every paid script with its settings, so an episode can be loaded again later instead of written again.

• Mage: scene scripts, filming, faces, voices, lip sync, the join.
• You: premise, outlines, cliffhangers, takes, captions, posting.
• Save the keepers: scenes are deleted after 30 days.