For many creators, the two biggest pain points in making AI short videos have never been “not having the right tools,” but:

Keeping the character’s look and style consistent

Controlling visual quality and overall polish

The result: you spend the time, but the final output still feels like a “half-finished demo.”

This article breaks down a battle-tested 5-step AI short-video workflow that compresses what used to take days—or even a full team—into just a few hours, while keeping characters consistent, visuals stable, and the whole process repeatable.

‍

Step 1: AI Super Screenwriter — From an Idea to an “Executable” Storyboard

A high-finish short video isn’t driven by inspiration—it’s driven by a storyboard script you can execute immediately.

The biggest wrong approach is: “Write me an AI short film story.”

Vague instructions like that only get you a bunch of text you can’t actually use.

The right approach: from the start, require the AI to output in shot list (storyboard) format, not a normal article.

A proper shot list must include three elements:

- Visual description: clearly define the character, setting, and visual focus

- Cinematography language: shot size (close-up / medium / wide) and pacing

- Voiceover lines: one set of lines per shot that can be recorded directly (about 5–8 seconds)

👉 Resource: Template 1 | AI Super Screenwriter Prompt

(Copy and paste this prompt, then insert your concept:)

You are a professional short-film director and screenwriter. Turn my story concept into a 45–60 second short-video storyboard script.

Your output must include:

1. One logline

2. A pacing structure of 3-second Hook / 20-second Wrong vs Right / 7-second CTA

3. A shot list (minimum 8 shots). Each shot must include:

4. Setting

5. Shot size (close-up / close / medium / wide)

6. Camera angle (eye-level / top-down / low-angle)

7. Character action and expression

8. Key props or on-screen text

9. Visual style keywords

10. Dialogue (if any)

Constraints: each shot should be 3–5 seconds; shots must be editable and coherent.

My story concept is: {paste your concept}

Target audience: Hong Kong / Cantonese audience; tone natural and fast-paced.

‍

Once you finish this step, what you have is no longer an “idea”—it’s a blueprint you can hand to AI for execution.

Step 2: Character Consistency Setup (Character Consistency)

Most AI short videos fail not because the visuals aren’t pretty, but because—

The character looks different in every shot, like they’re not the same person at all.

The solution: build a Character Sheet (a “character ID card”).

Core rule: first separate “what can change” from “what must never change.”

Non-negotiables (must be locked):

- Face shape and facial proportions

- Skin tone, hair direction/parting

- Age impression, overall vibe

- Fixed outfit colors or signature accessories

Variables (allowed to change):

- Facial expression, pose

- Background/setting, camera angle

- Lighting, emotional tone

Reference Library Recommendation

Prepare at least 6 reference images:

- Front view

- Left profile

- Right profile

- Half-body

- Full-body

- Natural smile

Tool-wise, you can use Nano Banana Pro or similar solutions to lock consistency using a “master character image + reference images,” without needing to train a model from day one.

👉 Resource: Template 2 | Character Sheet Prompt

Please create a “Character ID Card” for me, split into non-negotiable elements and variable elements.

The non-negotiables must be detailed enough to act as consistency anchors (facial proportions, hair direction, vibe, signature accessories).

The variables should list what is allowed to change (outfit, background, pose, emotion, camera).

Finally, output one “consistency anchor description” (80–120 Chinese characters) that I can paste every time I generate.

Character concept: {character setup}

Step 3: Cinematic Visuals (Images First, Then Video)

If you generate videos directly from text (Text-to-Video), the result is usually—uncontrollable.

A more stable approach is a two-step method:

1. Text-to-Image

- Use the visual descriptions from Step 1

- Add the locked character references from Step 2

- Generate one static keyframe per shot

Why this works: in the “still image stage,” you can fully control composition, lighting, and wardrobe details—rather than gambling on randomness.

2. Image-to-Video

- Use the approved keyframe as the starting frame

- Use tools like Kling

- Add only a small amount of controlled motion instructions

👉 Resource: Template 3 | Shot Prompt (Image/Video)

Based on the consistency anchor and shot requirements below, generate a single-shot visual description (usable for image or image-to-video).

Consistency anchor: {paste anchor}

Shot ID: {S01_SH03}

Setting: {office/street/cafe...}

Shot size & angle: {close-up/eye-level...}

Action: {what the character does}

Emotion: {nervous/confident...}

Style: {realistic/animation/cinematic...}

Constraints: do not change facial proportions or signature accessories; keep on-screen text minimal and clear (if needed).

Step 4: AI Voiceover + Lip Sync (Lip Sync)

After the visuals are done, the biggest giveaway usually comes from two things:

Audio and mouth movements.

Workflow

1. AI voiceover

Use tools like Minimax

Turn the storyboard voiceover lines into Cantonese narration

2. Lip sync (only for necessary shots)

Only process close-up shots where the character really needs to speak

Use Kling’s Lip Sync feature

AI analyzes the audio and drives mouth movement automatically

👉 Resource: Template 4 | Lip Sync Prompt

Rewrite the following dialogue into Cantonese lines that are “lip-sync friendly”:

Short sentences

About 8–12 characters per line

Add natural pauses (“…”)

Emotion tags (in brackets: calm/excited/helpless)

Original dialogue: {paste}

Step 5: AI Editing Acceleration + Assembly

By the final step, there’s not much technical difficulty left.

You already have:

Character-consistent video clips

Lip-synced talking shots

Full voiceover audio

Recommended tools: JianYing (剪映) or CapCut

Suggested process:

Align video and audio on the timeline

Add royalty-free BGM

Manually adjust subtitles (size, pacing, emphasis) — it usually looks better than fully automatic captions

‍

Conclusion: The Real Key Is “Control”

This workflow isn’t about how many AI tools you use. It’s about this:

At every step, you know exactly what you’re controlling.

With structured prompts + staged generation (images first, then video), you stop being dragged around by AI randomness—and start working like a director, clearly deciding:

What the audience should see

What they should hear

What they should remember

Now, copy the prompts above and start making your first AI short video that actually feels finished.

‍