From idea to a finished episode, with no crew and no camera
The hardest part of making short drama with AI is not the tools. Tools change every few months and any of them can be learned in a day. The hard part is the order of work: what has to be locked first so later decisions do not invalidate finished work.
This course lays that order out as six stages flowing one way. At each stage you will know its output, what must be locked before moving on, and which mistake most often forces people to start over.
One part here is rarely taught elsewhere: writing while thinking about cost. The script a producer likes best is not the most poetic one, it is the one that is cheapest to make, fastest to read, and needs the fewest revisions.
After this course you can:
Built by the Microdrama AI team from the production flow used to make episodes on this platform.
For creators who want to produce a series without a crew. No technical background or video editing experience is needed, but you do need a story that already has shape.
AI production feels like it can be done in any order. The strongest temptation is to make visuals first, because the results appear fast and are fun to look at. It is also the most common reason people start over.
The reason is simple: visuals depend on characters, characters depend on the script, and the script depends on the outline. Changing something upstream invalidates everything already made downstream.
| If you skip ahead | What has to be redone |
|---|---|
| Visuals before the character is locked | Every image, because the face shifts |
| Scripts before the outline is done | Episodes that do not connect to the previous one |
| Voice before the script is final | Every recording whose line changed |
| Assembly before a shot list exists | The whole assembly, because durations do not fit |
The output of this stage is a beat sheet with one event per line, exactly as in Short Drama Story Structure. If you did that exercise, this stage is already done.
AI helps here not by generating ideas but by stress testing the one you have. Ask the model to name five series with a similar premise, then ask what makes yours different. If the answer is thin, the premise needs sharpening before production.
| What must be locked | Why |
|---|---|
| Episode count | Determines how large the conflict has to be |
| The central conflict | Repeated in every AI request throughout production |
| The lead and what they want | The basis of every dialogue and visual decision |
| The antagonist and what makes them dangerous | Without this, act two runs out of material |
Short drama scripts are short, usually 150 to 250 words for a 90 second episode. Because they are short, the temptation to skip this stage is strong. Do not. The script is the cheapest place to discover that a scene does not work.
Format is not a school rule. Format is a communication tool: your script will be read by an editor, a director, and whoever generates the visuals. If the format confuses them, everyone loses time.
The directors are seated. MAYA enters carrying a folder.
DIAN
You still have the nerve to show up?
MAYA
I came to take what is mine.
Dian laughs. Maya sets the folder on the table.
MAYA(quietly)
This is the deed to the company.
The room goes silent.
| Cannot be filmed | Can be filmed |
|---|---|
| Maya walks in with a complicated mix of feelings | Maya walks in, pauses at the door, then keeps going |
| Dian feels threatened | Dian closes the folder in front of him |
| The mood turns tense | Nobody reaches for their water |
Separate dialogue from action clearly, because the two are processed differently at the next stage: dialogue becomes voice, action becomes shots. A script mixing them into one paragraph slows every later stage.
A shot list breaks each episode into the images you will generate. A 90 second episode usually holds 12 to 20 shots of 3 to 8 seconds each.
| No | Who is visible | Distance | What happens | Sec |
|---|---|---|---|---|
| 1 | Maya | Chest up | Stops at the door, takes a breath | 4 |
| 2 | Dian | Chest up | Turns, smirks | 3 |
| 3 | Maya | Chest up | "I came to take what is mine." | 3 |
| 4 | Maya's hands | Close | Sets the folder on the table | 3 |
| 5 | Dian | Chest up | His face changes | 4 |
Backgrounds almost never help the story on a phone, so wide distances should be spent sparingly. The face has to read, and a face taking up a fifth of the screen does not read.
Generate characters before scenes. A character locked early, with a face, hair, and wardrobe for each act, will be used hundreds of times. Making scenes first means every scene risks being thrown away.
Image models remember nothing between requests. Every image is made from scratch, so resemblance has to be forced through a reference rather than hoped for.
| What breaks resemblance | What to do instead |
|---|---|
| An angle far from the reference | Pick an angle close to the reference; swap the shot if needed |
| Very different lighting | Lock one lighting mood per location |
| Wide framing with a small face | Move up to chest up |
Voice decides whether an episode feels alive far more than people expect. Ordinary images with the right voice land much better than the reverse.
Lock one voice per character at the start and record the settings. Test with three lines of different weight: one neutral, one angry, one quiet. Many voices sound good neutral and fall apart when they have to be angry, and anger happens constantly in this format.
| Test | What you are looking for |
|---|---|
| Three lines of different weight | The voice holds together going up and going down |
| Two characters in one conversation | You can still tell them apart with your eyes closed |
| Line length against shot duration | No line is longer than the shot carrying it |
That last check gets skipped most and costs most. A line longer than its shot forces you to choose between cutting the line and slowing the story, and both are bad. Checking it at the script stage is far cheaper than at assembly.
Every idea has a price. A cost-aware writer gets the same emotion for a tenth of the budget, and that is a far more useful skill than the ability to write big scenes.
| Expensive | Cheap, same emotion |
|---|---|
| A plane crash | A father failing to arrive at his daughter's wedding |
| A five car chase | One incoming message containing a home address |
| Fifty extras in a hall | Three people in a lift who stop talking |
| A warehouse explosion | The warehouse key thrown into a drain |
Raise the emotion before you raise the budget.
Assembly combines shots, voice, music, and text into a finished episode. Before releasing, run the same checklist every time so nothing slips past you simply because you have grown too familiar with the material.
Generating twenty good scene images, then deciding what the lead's face looks like.
Locking one face reference per character, then generating scenes.
Image models remember nothing between requests. Deciding the face later means those twenty images will not match a single later episode.
Action line: "Maya walks in with feelings that are hard to describe."
Action line: "Maya walks in, pauses at the door, then keeps going."
Feelings cannot be generated or filmed. What can be executed is action, and the action is what conveys the feeling anyway.
A twenty-five word line written for a three second shot.
The line cut to eight words, or the shot extended to six seconds in the shot list.
This mismatch always surfaces at assembly, exactly where fixing it costs most. At the script stage it costs nothing.
A highway chase to show the lead is cornered.
One incoming message containing her mother's home address, no other text.
The emotion is equal or stronger, the cost is near zero, and on an upright screen a phone message reads far better than a fast-moving car.
A series using twenty different locations across eighty episodes.
A series reusing six locations, with one new location marking the start of act three.
Reused locations build a sense of place and cut consistency cost. A new location then means something precisely because it is rare.
One finished episode plus a six row timing table.
A shot list whose total duration already fits.
A before and after table for locations, characters, and props.
You are ready to move on when all of these hold:
For this workflow, no. Assembly can be done with tools that combine shots, voice, and text without advanced editing. Editing skill helps, but it is not a requirement.
Because visuals depend on characters and the script. Making them before both are locked means every story change throws away finished images.
It depends heavily on experience and tools. More useful than an average is recording your own times in the first exercise and using them to schedule the full series.
