How to Make an AI Micro Drama (Costs + Workflow)

Aug 1, 2026

Making an AI micro drama is four steps — lock the format, build character references, render one shot at a time, assemble — and three numbers: 18 shots an episode, 60 episodes a season, and roughly $18 to $73 per episode depending on the tier. The rest of this page is those steps and where each one breaks. Last updated August 2026.

TL;DR — An AI micro drama is a 9:16 vertical series of 60–120 second episodes with a cliffhanger on every one, and a season is typically 60–100 of them. At five seconds per generated shot, a 90-second episode is roughly 18 renders. On the cheapest tier here that is about 1,350 credits (≈ $18) per episode, or ≈ $1,090 for a 60-episode season. The bottleneck is not writing and it is not rendering — it is keeping one face the same across 18 shots, which is why reference-image models matter more than raw video quality.

What an AI micro drama actually is

The micro drama format came out of China's duanju boom, and it is far more rigid than "short video". Get the constraints wrong and a micro drama reads as amateur regardless of render quality.

Constraint The norm Why it matters
Aspect ratio 9:16 vertical Watched one-handed, full-screen, on a phone
Episode length 60–120 seconds Long enough for one beat, short enough to autoplay into the next
Season length 60–100 episodes The model is a binge, not a single hit
Structure Cliffhanger on every episode Retention is what the format is engineered for

The scale is not niche: the Chinese micro drama market passed 50 billion yuan (about $6.9 billion) by 2024 with roughly 662 million domestic viewers, which is why "drama" and "story" modes started appearing across AI video tools.

The one structural rule

One micro drama episode carries exactly one turn — a reveal, a betrayal, a reversal — and ends inside the next one. Writing three beats into a 90-second episode is the most common beginner mistake, because it leaves nothing to cliffhang on.

The five decisions before you generate a micro drama

Make these five before writing a single micro drama prompt. Changing any of them later means re-rendering everything.

  1. Shot length. The tiers here generate in fixed slots — 5 seconds on Lite, 5 or 10 on Fast, 5/10/15 on Turbo. Five seconds is the safer default: a shorter clip has less room to drift away from your reference, and costs less to throw away.
  2. Character count. Every additional recurring face multiplies your consistency problem. Two leads and one antagonist is plenty.
  3. Location count. Backgrounds drift as badly as faces. Three or four locations reused across the season keeps continuity cheap.
  4. Audio strategy. Models that generate synchronised audio in the same pass — Omni generates dialogue and effects alongside the video — cost more per second than a silent render, but save you assembling audio separately.
  5. Episode count. Decide the length of the micro drama up front and multiply by your per-episode cost before you start, not after episode three.

The five-stage AI micro drama production pipeline: script, character reference stills, shot list, image-to-video render, assembly — with the reference stage marked as the expensive one to get wrong
The render step is the cheapest to redo. The reference step is the most expensive to get wrong.

Keeping one character consistent across a micro drama

This is the hard part of AI micro drama production, and it is worth more of your effort than the script.

Cost of one 90-second AI micro drama episode across five model tiers, from 1,350 credits on Kling 4.0 Lite to 5,400 on Omni with audio
Eighteen five-second shots per episode. Tier choice moves the number four-fold.

In a micro drama, a text prompt describing a person — "woman, 30s, dark bob, red coat" — will produce a different woman every render. The fix is reference-based generation: you supply actual images of the character and the model matches them instead of re-inventing from words. Kling 3.0 Omni accepts up to seven reference images per generation for exactly this, and also takes existing video as input so a shot that worked can seed the next one.

The practical workflow

  1. Generate your cast as stills first, on the image generator. Iterate here — a still costs a fraction of a clip.
  2. Lock three or four angles per micro drama character: front, three-quarter, profile, and one full-body for wardrobe.
  3. Feed those as references into every shot the character appears in, for the entire micro drama, not just the first episode.
  4. Re-roll aggressively. A shot that is 80% right will be visibly wrong next to the other seventeen.

Micro drama locations get the same treatment: one establishing still per location, reused as a reference forever.

What an AI micro drama episode costs

Here is the per-episode micro drama number, computed from this site's live credit rates. Assume a 90-second episode built from eighteen 5-second shots, priced at the Starter pack rate of $19.90 for 1,480 credits (about $0.0134 per credit).

Tier Per 5s shot Per 90s episode ≈ Cost 60-episode season
Kling 4.0 Lite 75 credits 1,350 $18.15 ≈ $1,089
Kling 4.0 Fast (1080p + audio) 95 credits 1,710 $23.00 ≈ $1,380
Kling 3.0 Turbo (std) 110 credits 1,980 $26.62 ≈ $1,597
Kling 3.0 (std, no audio) 175 credits 3,150 $42.35 ≈ $2,541
Kling 3.0 Omni (std, audio) 300 credits 5,400 $72.61 ≈ $4,357

Two things fall out of the micro drama table. First, the tier choice is a four-fold swing — the same season is roughly $1,100 or roughly $4,400 depending on nothing but which model you pick. Second, re-rolls move the number more than anything else you control. Assume you render each shot twice on average and every number above doubles; the discipline of nailing the reference stills first is worth more than any prompt trick.

A reasonable micro drama strategy: draft the whole episode on the cheapest tier to check pacing and continuity, then re-render only the shots that carry the emotional beat on Omni with audio. Full plan and pack pricing is on the pricing page.

Where AI micro drama production still breaks

Honest micro drama limitations, because they will shape what you write:

  • Dialogue-heavy scenes. Sustained lip sync is where generated video is most likely to give itself away. Write for reaction shots, voiceover, and short lines.
  • Physical continuity. A cup on the left in one shot lands on the right in the next. Compose around it — avoid props that must persist.
  • Crowds. Background people mutate between shots. Keep scenes to your named cast.
  • Hard cuts within a shot. Each generation is one continuous take, so every cut in your micro drama is a separate render. Cost still tracks total runtime on the per-second tiers — but Lite caps a shot at five seconds, so a cut-heavy episode is locked to that granularity there.
  • Long emotional holds. Five seconds is not enough for a slow realisation. Break it into two shots with a cutaway between.

Write the script knowing all of this and micro drama production gets dramatically easier. Write a conventional screenplay first and you will spend the budget fighting the tool.

FAQ

Are micro dramas AI-generated?

The commercial industry that produced those $7 billion in revenues is built on live-action shoots, not generated video. Fully AI-generated series are a much newer category — which is precisely why the format is still open to independent creators.

How do you create AI short dramas?

Lock the micro drama format first (9:16, 60–120 seconds, one turn per episode), generate your cast as reference stills, then produce each episode as roughly eighteen 5-second image-to-video shots using those references. Budget by shot count, not by runtime.

Is it legal to make AI movies?

This is not legal advice, and the answer depends on your jurisdiction. The risk concentrates in the inputs rather than the technique: a real person's likeness and a protected character or franchise design are the two things most likely to cause a problem, and several platforms now require AI-generated content to be labelled. Original characters and original scripts are the low-risk path.

How is an AI micro drama made in practice?

Script → character reference stills → shot list → image-to-video render per shot → assembly and audio. The render step is the cheapest part of that chain to redo, and the reference step is the most expensive to get wrong.

How long does one episode take to make?

Less than you would guess for the rendering and more than you would guess for everything before it. The reference stills and shot list are the slow part, and they are reusable across the whole season — so episode one costs far more effort than episode twenty.

Can I make an AI micro drama for free?

Not a whole one. New accounts get 100 free credits, which covers a single 5-second clip — enough to see how a model handles your character reference, not enough to finish an episode. The arithmetic above is the honest floor.

Sources

Probiere es direkt aus

Erstelle ein Video mit der Kling-Flaggschiff-Engine — 100 Gratis-Credits bei der Registrierung, keine Karte nötig.