AI Short Drama Prompt QA: 12 Checks Before You Submit a Shot for Rendering

Maosika Editorial ｜ Last updated 2026-08-03

Before an AI shot is sent to render, the prompt should read like something a director can approve—not a paragraph of mood words. This 12-point QA checklist turns prompt review into a repeatable handoff step for vertical short drama teams.

Treat prompt review as a handoff, not a polish step

In live-action production, a shot does not roll until the call sheet, set, cast, props, wardrobe, and shot description are aligned. AI short drama production needs the same discipline. The prompt is not creative brainstorming text; it is the production instruction that reaches the video model at render time.

A shot prompt QA pass is the final review before a field is submitted for rendering. Its job is to catch ambiguity, continuity drift, reference conflicts, impossible camera moves, and missing constraints while they are still cheap to fix.

This checklist is written for directors, editors, prompt operators, and production supervisors working on vertical micro-dramas. It assumes the script has already been broken into short shot blocks, character and scene references have been prepared, and the team is now deciding whether each shot is ready to render.

The 12-point shot prompt QA checklist

Use the table below as the final gate before clicking render. If an item fails, send the shot back for rewrite or asset preparation rather than hoping the model will infer the right answer.

1. Subject clarity: name who is in the shot

The first thing to check is whether the prompt identifies the correct subject. In a short drama, “the man” or “the girl” is not enough when several characters share similar traits.

Use the character’s production name and make sure it maps to the correct approved reference. If the script calls a character “Officer” but the cast bible uses a full name, bind the role to the right character before rendering.

A useful question: Could a stranger on the crew pick the right actor from this prompt? If not, rewrite it.

2. Action specificity: write what the camera can see

AI video models render visible behavior, not internal psychology. A prompt that says “she realizes the truth” gives the model nothing concrete to film.

Replace internal states with visible actions:

• “She freezes, eyes widening.”

| “He grips the letter until his knuckles whiten.”

| “The door opens before she can finish speaking.”

The action should answer: what exactly changes in the frame during this shot?

3. Action scale: prefer one controlled movement

Vertical drama shots are usually short, often around 10 seconds per shot block. That is not enough screen time for several large actions to read clearly.

A strong prompt usually gives the subject one primary movement. For example: standing up slowly, reaching for a phone, turning toward the door, or lowering a weapon. Complicated action sequences should be split into multiple shots rather than compressed into one.

This does not make the drama less dynamic. It makes each beat readable.

4. Set and time: lock interior/exterior, location, and day/night

Scene continuity begins with basic production information. The prompt should make clear whether the shot is interior or exterior, where it happens, and whether it is day or night.

This information should come from the structured script breakdown, not from improvisation. If one shot in a scene says “modern office at night” and another says “daytime apartment,” the scene will not cut together.

5. Props: include what matters, exclude what does not

Props should be tied to the action. A mobile phone matters if the character reads a message on it. A knife matters if it is revealed, picked up, or used as a threat. Unrelated props add visual noise.

Before rendering, confirm that:

• required props are listed in the script’s original wording where possible,
• prop reference images are ready if needed,
• irrelevant props are excluded from the shot,
• the prop does not contradict an earlier episode.

6. Reference-image rule: do not overwrite the approved look

This is one of the most important checks in AI short drama production.

If a character has an approved reference image, the prompt should not re-describe that character’s clothing, hairstyle, or facial features. The reference image is the authority. Text should focus on action, expression, and temporary changes such as wounds, dirt, or tears.

When the prompt repeats wardrobe details, the model receives two competing sources: the image and the text. That conflict often causes outfit changes, face drift, or inconsistent styling.

A simple rule for the reviewer:

Referenced face and wardrobe: use the image. Unreferenced action and condition: use words.

7. Lighting and tone: make the scene cut together

Lighting is not decoration. It tells the audience whether two shots belong in the same scene.

Check that the prompt specifies enough lighting information for continuity:

• light source direction,
• time of day,
• hard or soft shadow quality,
• warm/cool color tendency,
• practical lights such as lamps, neon, phone screens, or candles.

A cliffhanger can be dark, but it still needs a readable face and a clear visual focus.

8. One camera move per shot

A common prompt mistake is stacking camera instructions: “push in slowly, then pan left and tilt up.” That asks for several decisions at once and often produces unstable motion.

Choose one camera intention per shot:

• slow push-in for realization,

| lateral track for tension,

| static hold for dialogue,

| slight pull-back for revelation,

| low-angle push for power.

If the story needs more movement, split it into separate shots in the edit plan.

9. Shot language: specify size, angle, and composition

The prompt should communicate the shot’s visual grammar. The team does not need to write a cinematography essay, but it should identify the basics:

• shot size: close-up, medium shot, wide shot,
• angle: eye-level, low angle, high angle,
• composition: centered, over-shoulder, profile, foreground obstruction,
• subject placement for vertical 9:16 framing.

Vertical screens reward clear composition. If the face is the story point, do not bury it in a wide, busy frame.

10. Style anchor without IP names

The visual style should be consistent across the production. However, prompts should not rely on named films, shows, or third-party IP. That can create legal or platform-filter risk and does not give the model stable technical instructions.

Instead, anchor style through controllable visual language:

• rendering family: 2D, 3D, realistic, illustrative,

| line quality and brush texture,

| skin and material rendering,

| color palette,

| lens texture,

| contrast level.

The style guide should do the aesthetic work; the prompt should carry the shot-specific instructions.

11. Constraints and safety: add the guardrail package

Even a beautiful prompt can fail without constraints. Before submission, confirm the prompt includes the necessary guardrails for the shot type.

Typical constraints include:

• stable face and identity,

| no watermark, logo, or text artifact,

| no duplicated faces in multi-character shots,

| no extra fingers or distorted hands when hands are visible,

| clean continuity of props and wardrobe,

| style consistency with the approved look.

These constraints are not creative limitations. They are production quality controls.

12. Render settings: do not approve the wrong deliverable

A prompt can be perfect and still produce the wrong asset if the render settings are incorrect.

Before submitting, verify:

Render settings are part of creative approval. A shot generated in the wrong ratio is not “fixable later” without compromising composition.

A fast review workflow for the field

For a high-volume short drama production line, prompt QA should be quick but standardized. A practical workflow is:

1. Read the script beat. Confirm the shot’s dramatic purpose.
2. Check assets. Confirm character, set, and prop references are available or intentionally omitted.
3. Read the prompt aloud. If it sounds like a novel, rewrite it.
4. Mark the camera move. There should be one clear intention.
5. Scan for appearance contradictions. Remove wardrobe/face text when references exist.
6. Confirm constraints. Add stability, no-logo, and duplicate-face guardrails as needed.
7. Check render settings. Verify ratio, duration, tier, resolution, and audio.
8. Submit as one take. If the result fails, edit the prompt or references and generate a new take.

The goal is not to make every prompt perfect on the first render. The goal is to make every failure diagnosable. If a shot breaks, the team should know whether the problem came from action, reference, camera, lighting, style, or settings.

What this checklist does not solve

A checklist cannot fix a weak story, a missing look-development process, or inconsistent character art. It also cannot replace human review of the final take.

There is no universal automatic scoring system that can decide whether a shot is dramatically good. The final judgment still belongs to the director, editor, or producer. The checklist simply prevents avoidable production errors from reaching the render queue.

About Maosika

Maosika is an AI production operating system for vertical short dramas. It structures the workflow from idea to locked brief, script batches, story archive, look selection, character/scene/prop art, shot blocks, reference mapping, multimodal rendering, review, and retakes. Its position is that repeatable, drift-prone production steps should become a controllable pipeline, while aesthetic judgment remains with the creator.

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