Create a Short AI Video

Start with an example

AIH’s own guide, R.I.C.K. Bot, was made with the techniques in this lesson. Each clip was planned like a single camera shot:

Who: R.I.C.K. Bot, a friendly navy-and-gold robot with cyan eyes, holding a glowing tablet.

What happens: He raises his hand and waves at the camera, then speaks.

Camera: Slow push-in at eye level.

Light: Bright studio with a soft navy-to-white background.

Sound: He says, “Hey, I’m R.I.C.K. Bot, your guide at Applied Intelligence Holdings. Glad you’re here.”

Every clip started from the same set of reference images, so he looks like the same character in all of them. Four clips edited together make a 30-second greeting.

How AI video models work

Video models build on the diffusion idea you learned in Lesson 1, with a big extra challenge: time.

A video is a series of still frames, often 24 per second. An 8-second clip is about 200 images that must:

• Show the same subject with the same face, clothes, and colors in every frame
• Keep the lighting and background consistent
• Create believable motion from one frame to the next
• Match the sound, such as lips moving with dialogue, in tools that generate audio

That’s why video takes longer and costs more than images, and why it can show glitches: a hand that changes shape mid-gesture, an object that appears from nowhere, or text that shimmers. The model is predicting motion patterns, not following the laws of physics.

Think like a director

AI video tools usually generate clips of about 4 to 15 seconds. Plan your video as a set of shots, and give each one a clear description:

Part	Describe	Example
Who	The person, character, or product	“A smiling baker in a flour-dusted apron”
What happens	One clear action	“Pulls a tray of croissants from the oven”
Camera	How we see it	“Slow push-in, eye level”
Light and style	The look	“Warm early-morning light, realistic”
Sound	Dialogue, music, or effects	“She says, ‘Fresh every morning at six.’”
Format	Shape	“Vertical 9:16”

One action per clip. “She pulls the tray out, sets it down, waves, walks to the counter, and hands a croissant to a customer” is five actions. The model will rush, blend, or skip them. Split it into separate clips.

Camera movement vocabulary

Movement	What it does	Feeling
Static	Camera doesn’t move	Calm, focused
Push-in (dolly in)	Moves closer to the subject	Builds attention or emotion
Pull-out	Moves away	Reveals the setting
Pan	Turns left or right	Follows action or shows a space
Orbit	Circles the subject	Shows off a product or character
Tracking shot	Follows a moving subject	Energy, journey
Handheld	Slight natural shake	Real, documentary, user-generated feel

Dialogue and timing

Many video tools can now generate speech. Keep lines short:

• People speak about 2 to 3 words per second.
• An 8-second clip fits about 16 to 20 words comfortably.
• Put the exact line in quotation marks: He says: “Fresh every morning at six.”
• If the voice changes between clips, consider recording or generating the voiceover separately and adding it during editing.

Two ways to start

1. Text to video: Describe the whole shot in words. Fastest, but least control over how the character or product looks.
2. Image to video: Start from a still you’ve approved, from Lessons 1–3, and describe only the motion. This is the best way to keep a character, product, or logo consistent, and it’s how professional AI creators work.

Video tools include Google Veo, OpenAI Sora, Runway, Kling, and others, often available inside other apps. Plans, limits, and prices change often, so check what your account includes.

Save money: still first, then motion

Video generation uses many more credits than images. A smart, professional workflow:

1. Design the scene as a still image and refine it until it’s right.
2. Animate that still with a shot description focused on motion and camera.
3. Fix with small changes: “Same shot, but a slower camera” or “Same scene; she smiles at the end.”
4. Approve each clip before moving to the next.

From clips to a finished video

A finished ad or explainer is built in an editor, such as CapCut, Canva, Adobe Express, or iMovie:

1. Arrange your clips in order.
2. Trim the start and end of each clip to remove awkward frames.
3. Add captions. Most social video is watched without sound, so captions matter.
4. Add music, your logo, and on-screen text. Don’t ask the AI to generate text inside the video, because it often garbles it.
5. Export in the right format for each platform.

Common problems and fixes

Problem	Fix
Character looks different in each clip	Start every clip from the same reference image
Too much happening	One action per clip
Weird hands or faces	Use a wider shot, simplify the action, or regenerate
Voice changes between clips	Generate or record the voiceover once and add it in editing
Text is garbled	Add text later in an editor
Motion looks frozen or dull	Add a specific camera move and a clear action verb

A worked example: a 3-clip ad

Carlos owns a small bike repair shop and wants a 24-second Instagram Reel. His shot plan:

Clip	Action	Camera	Line
1. Hook	A frustrated teen looks at a flat bike tire on a sidewalk	Slow push-in	“Flat tire, and practice is in an hour.”
2. Solution	Carlos swaps the tube at his workbench, then spins the wheel	Orbit around the bench	“Walk in, ride out. Most fixes take 20 minutes.”
3. Call to action	The teen rides away down a sunny street	Tracking shot	“Carlos’s Cycles, on Milwaukee Avenue. Open till 7.”

He makes a reference still of himself at the workbench first. Clip 2 comes out with his hands blurring during the repair, so he simplifies the action to “spins the repaired wheel and gives a thumbs up” and regenerates. He edits all three clips in CapCut, adds captions and his logo, and labels the post as AI-generated, as his platform requires.

Be honest

• Label realistic AI video when you post it. Major platforms, including YouTube, TikTok, Instagram, and Facebook, require labels for realistic AI content.
• No fake testimonials or fake customers. The FTC treats fake reviews as deceptive, and its rules allow penalties for them.
• Don’t recreate real people’s faces or voices without their written permission.

Think deeper: why video is so much harder

An image only has to look right once. A video has to look right about 200 times in a row. An eight-second clip at 24 frames per second is nearly 200 images, and every one has to match the one before it. A face can’t change shape. A shirt can’t change color. That challenge is called consistency, and it’s why video costs so much more than images.

Video models learned what motion usually looks like, not the laws of physics. They’ve seen millions of people walking and liquids pouring, so common actions look great. Unusual ones, like hands doing something complicated or objects passing behind other objects, can glitch. Keeping actions simple, clips short, and tricky details off-camera all help.

Keeping a character the same across clips is a separate problem. Each clip is generated fresh, so professionals use a character sheet: approved reference images from several angles plus the exact same description every time. Cartoon studios have done this for decades so that many artists draw a character the same way.

Questions to sit with:

• An AI video can look real without the AI understanding gravity, cause, or effect. What does that suggest about the difference between looking right and being right?
• As AI video gets more realistic, what will you need to see before you believe a video is real?

Project

Plan and create a 3-clip mini ad: a hook, a solution, and a call to action. Write a shot plan for each clip, create a reference still to keep your main character or product consistent, generate each clip, improve at least one with a follow-up, and edit the clips together with captions in a free editor.

R - Role: You are a film director making a short social video.
I - Instructions: Create an 8-second clip of [who] [doing what].
C - Context: Setting: [where]. Style: [realistic / 3D animated / cinematic]. Mood: [upbeat, calm]. Camera: [slow push-in / static / orbit / follows the subject]. Lighting: [golden hour, studio, neon].
K - Key Output: [Vertical 9:16 / wide 16:9]. Sound: [the person says: "..." / upbeat music / room sounds]. No on-screen text.