How AI Vertical Short Dramas Actually Get Made: A Production-First Workflow From Idea to Final Cut

Maosika Editorial ｜ Last updated 2026-07-25

Most AI short drama failures are not model failures. They are order failures: teams try to render before the brief is locked, shoot before characters are designed, and write Episode 20 before Episode 3 is archived.

The workflow problem no model can fix

When a vertical short drama goes wrong, the symptom usually appears on screen: faces change, costumes drift, props disappear, cliffhangers do not land, and Episode 12 contradicts Episode 2. But the root cause is usually earlier in the pipeline.

AI generation tools are very good at producing one impressive clip. They are not naturally good at producing 60 or 80 clips that belong to the same show, with the same characters, the same rules, the same debts owed to earlier episodes, and the same release rhythm.

That gap is what a production workflow is for.

A production workflow is the enforced order in which decisions must be made, reviewed, locked, and passed downstream. It exists to prevent the team from spending render budget on scenes that will be rewritten, or asking a video model to invent a character who has never been designed.

The industrial order, from idea to final cut

The order matters more than the tool stack. If you rearrange the steps, quality collapses even if every individual generation looks good.

This is not a software menu. It is the order a live-action crew would recognize: development first, then script, then casting and art, then shot planning, then production, then editorial.

Stage 1: Intake and creative assessment

The first job is not to write. It is to discover whether the idea is complete enough to write.

A useful intake covers the vertical-short-drama dimensions that actually determine production decisions:

• Genre and subgenre
• Protagonist and character arc
• Core conflict
• Story direction and ending direction
• Episode count and per-episode length
• Tone and visual baseline
• Hook rhythm and satisfaction beats
• Target platform and audience

For vertical dramas, per-episode length usually lives in the 1–2 minute range, with an outer ceiling around 3–4 minutes. That length constraint changes writing: there is no room for slow setup.

A healthy system should score intake completeness and branch accordingly. If the brief is nearly complete, it should move forward quickly. If it is thin, it should force the missing questions to be answered before writing begins. In production terms, this is the development meeting: nothing shoots until the brief is locked.

Stage 2: The creative brief is a lock, not a suggestion

The creative brief is the document every downstream stage treats as law.

A locked brief should include:

• A one-sentence logline
• The central contradiction driving the show
• The overall story direction
• The ending direction
• Satisfaction beats and hook rhythm
• Target platform and audience
• Episode-by-episode outline
• Specific instructions for the writer
• Character arcs for the leads

The character arc requirement matters. A lot of AI-written leads are just people to whom things happen. A lead with an arc has a starting state, a pressure point, a transformation, and a cost.

Once the brief is locked, the next stage can begin. Before that, every asset produced is speculative waste.

Stage 3: Build the continuity archive before writing long

Continuity archive is a structured record of everything that must remain true across the whole drama.

It contains:

• Character identities and stable traits
• Changeable current state: injuries, hidden identities, alliances, possessions
• Relationship maps
• Open and resolved plot threads
• Episode-by-episode appearance records
• Batch-by-batch plot summaries
• Visual descriptions of recurring props

This is the AI-era replacement for relying on model memory. Models do not reliably remember what happened 30 episodes ago, especially across long sessions. A structured archive does.

When a new batch of episodes is written, the writer should only consume the relevant archive slice: current character states, unresolved threads, recent batch summaries, and the current batch objective. After the batch is approved, the archive updates.

That is how a show stays coherent at scale.

Stage 4: Write in batches, with beat sheets before prose

Vertical short drama writing has its own internal discipline. It is not prose fiction, and it is not traditional TV pacing compressed.

The rules that reliably work:

1. Golden 3 seconds: the first scene must open with conflict or a hard question. No soft setup, no background lecture.
2. Episode shape: opening hook, escalation, final-scene cliffhanger.
3. Beat density: at least one small satisfaction beat per episode; a larger payoff every few episodes.
4. Dialogue: short lines, usually under 20 characters in Chinese-language originals and similarly tight in English adaptation; no essay speeches, no explanatory narration.

Before drafting any episode, the writer should produce a beat sheet listing the scenes and the cliffhanger. Only after that is approved should the episode prose be written.

From the second batch onward, four things should be confirmed before writing: the batch direction, the focal characters, the threads and conflicts to advance, and the end-of-batch hooks. Those become hard constraints. If they conflict with older archive notes, the newly confirmed creative intent wins.

Every finished batch should pass a rule-based quality check before delivery: correct episode titles, matching scene counts, required character lines, enough dialogue, no placeholder text like "to be continued." If it fails, it goes back for rewrite with the specific failure attached.

Stage 5: Casting, look development, and the reference image discipline

This is where most AI productions lose control.

Characters must be designed before shooting. That means final character art, scene art, and recurring props, all under one chosen visual style. A show should have a defined style manual covering character rendering, environments, props, and video style tags, so the lead does not look like 2D anime in Episode 1 and realistic footage in Episode 4.

There should be a hard gate before shooting: cast is complete, names are valid, visual fields are complete, and leads match the brief.

The most important consistency rule is simple:

If a character has a reference image, the prompt must not re-describe that character's clothing or appearance in text. The image owns the look. The text only describes action, expression, and injury.

This rule directly prevents the classic AI failure where text and reference fight each other and produce a different face or outfit every scene.

For each scene, reference assets should be assembled in a fixed priority: scene image first, then props for that scene, then character designs. If an asset does not exist, the slot should be marked as text-only rather than silently invented.

A few practical controls matter here:

• Manual character binding when the script uses a title or nickname
• Manual prop inclusion or exclusion to control visual focus
• Version swapping when an older design works better for a specific scene
• Era-based costume routing for time-slip or flashback stories
• Empty scene plates with no people in them

Reference images are not a forced gate. You can skip them and shoot from text. But the professional path is to finish lookdev first, because skipping it usually costs more in retakes later.

Stage 6: Cut episodes into scene blocks before shooting

A finished episode should not be sent to video generation as one long prompt.

Instead, each episode is cut into scene blocks, roughly 10 seconds each, with a soft cap around 200 characters of script body. Longer action is split by beats, punctuation, or paragraph breaks. Crowd characters and generic extras are separated from the named cast so they do not consume design slots.

This is the production equivalent of seeing a clip list on an editing timeline: Episode 3 becomes Scene 1, Scene 2, Scene 3, and so on. Each scene has its own prompt, its own references, its own output, and its own history of takes.

The default deliverable is vertical 9:16 from the start, not a horizontal video cropped afterward. That changes composition, staging, and how close the camera can be.

Stage 7: Write shot prompts like a shot list, not a novel

A strong video prompt is engineering, not poetry.

The eight core elements are:

1. Precise subject
2. Action detail
3. Scene environment
4. Lighting and color tone
5. Camera movement
6. Visual style
7. Image quality constraints
8. Negative or stability constraints

For simple scenes, one paragraph is enough. For complex cinematic scenes, a three-part structure works better: overall setup, shot-by-shot instructions, then a constraint package.

A few hard prompt rules prevent common failures:

• One camera move per shot; do not stack push, pull, pan, and tilt together
• Use shot numbers, not absolute timestamps
• Include a base quality package: face stability, no watermark, clean rendering
• Add twin or duplicate prevention in multi-character scenes
• Keep actions low, continuous, and physically clear rather than explosive and ambiguous
• Mark dialogue, sound effects, and music with consistent notation
• Feed only that scene's assets into the prompt; do not leak props or cast from other scenes

The goal is to make the model read a shot list, not improvise a paragraph.

Before delivery, prompts should also be cleaned of specific copyrighted IP names while preserving technique and aesthetic intent. If a reference image is identified as a real-person photo, the workflow should warn the creator to use platform-generated art instead of forcing a real face through the video pipeline.

Stage 8: Shooting, retakes, and the honest role of the human

The shooting stage should feel familiar to anyone who has worked on set.

For each scene, the creator reviews the prompt, checks references, chooses model tier and resolution, sets duration, submits the shot, waits in queue, reviews takes, and decides whether to keep, rewrite, rebind, or reshoot.

The controls a director would care about are:

A reliable production queue should behave like production software, not a toy: separate task lanes, no duplicate submissions on the same scene, pre-charge and release on failure, timeout recovery, retryable failures, and actual measured duration rather than trusting vendor metadata.

There is one thing it should not pretend to do: automatically judge which take is best. Final quality judgment stays with the creator. The system provides multiple takes and the tools to improve them; it does not replace the director's eye.

Stage 9: Assembly is still a separate step

The output unit is one scene, not one finished episode.

That means final editing remains a real stage: ordering the clips, trimming, pacing transitions, adding sound, and exporting the episode. The ~10-second scene block is a production heuristic, not a precision timecode cut, so some blocks may run long and need editorial adjustment.

There is currently no mature one-click workflow that automatically extends or continues video across scenes in a reliable way. The professional approach is to control each scene, then assemble.

Where AI helps, and where it does not

The honest boundary matters.

AI can enforce structure, maintain archives, generate first drafts, produce art under style rules, build shot prompts, render many takes, and keep the pipeline moving. It cannot guarantee a hit, replace taste, or automatically know which take is the right one.

A production-grade AI system does not promise zero human work. It turns the repetitive, drift-prone, failure-prone parts of the process into a constrained pipeline, while keeping审美 judgment in human hands at the checkpoints that matter: brief lock, character approval, prompt review, take selection, and final cut.

High-quality short drama is never one giant generation cut into pieces. It is many controllable units, built in the right order, assembled into a show.

If you are looking for this exact production order built into software, Maosika (猫斯卡) is designed around it: an AI production operating system for vertical short dramas that locks the brief, maintains the continuity archive, manages lookdev, blocks scenes, generates shot-list-style prompts, and supports multi-take shooting under human review.

About Maosika — Maosika · Professional AI Video Production System. It connects briefing, scripting, look development, shot prompts and delivery into one reviewable pipeline. www.maosika.com