How do you control an AI video camera? Shots, subject motion and keyframes matter more than style words

Lesson 017 established a key difference between image and video prompts: video instructions must describe time.

A second problem appears quickly in practice: the scene is attractive, but the camera seems to have no plan.

Before adding more words such as cinematic, epic or masterpiece, separate the responsibilities inside the shot.

Use a simple shot structure

A practical structure is:

Shot / composition
+ Subject motion
+ Camera motion
+ Environment / lighting
+ Duration

For example:

Medium shot of a woman sitting by a train window.
She slowly looks up from a book and turns toward the glass.
The camera makes a gentle push-in. No pan or rotation.
Warm sunset light, subtle train vibration, 6 seconds.

Each sentence tells the model something different.

Shot size sets the starting relationship

Common shot sizes include:

• Wide shot — emphasizes the environment.
• Medium shot — balances a person and their action.
• Close-up — emphasizes a face or detail.
• Extreme close-up — isolates a very small detail such as an eye or product texture.

If you need both an establishing wide shot and a facial close-up, two clips are often more controllable than asking a six-second generation to perform a complex edit by itself.

Subject motion and camera motion are different

Compare:

Subject: walks toward the doorway
Camera: locked on a tripod

with:

Subject: stands still
Camera: slowly dollies toward the subject

The person becomes larger in frame in both cases, but the physical event is different. Explicitly stating what moves can reduce ambiguity.

Learn a small camera vocabulary

A beginner does not need an entire cinematography textbook. A few terms already help:

• Pan — rotate left or right from a fixed position.
• Tilt — rotate up or down.
• Dolly / push-in — move the camera physically closer or farther.
• Tracking / truck — move laterally with a subject.
• Orbit — move around a subject.
• Handheld — intentionally less stable motion.
• Static / locked-off — no camera movement.

More movement is not automatically better. A clear subject action plus one gentle camera move often looks more deliberate than several simultaneous moves.

Do not force an entire scene into one clip

A prompt such as this is overloaded:

She enters, sits down, picks up a cup, drinks, stands, walks to the window,
while the camera orbits, zooms out and ends on an aerial city view.

Break it into shots instead:

Shot 1: enter the room
Shot 2: sit and pick up the cup
Shot 3: close-up of the drink
Shot 4: walk to the window

Editing is designed to combine shots. You do not need one generation to perform the entire film.

Keyframes anchor important visual states

Lesson 037 introduced first and last frames. These are examples of keyframes: important visual states you want to constrain.

If the start must show a closed door and the end must show the door open with a person outside, frame constraints can be more reliable than hoping the text alone produces the exact ending composition.

Duration changes perceived speed

Duration is not only a file length. It determines how much time the action has to unfold.

“Stand up” over three seconds feels different from the same action over ten seconds. Write timing deliberately when the tool exposes duration controls.

A useful iteration order

1. Fix the shot size.
2. Give one primary subject action.
3. Give one camera movement.
4. Check whether the motion is physically plausible.
5. Add lighting and style.
6. Add complex transitions only after the basic shot works.

This order makes failures easier to diagnose.

One thing to remember

A strong AI video prompt is a shot plan, not a bag of style adjectives. Define the composition, subject motion, camera motion and time window first; then decorate the image.