Direct the camera: shot sizes and camera moves for AI video
The words AI video models understand for shot sizes, lenses and camera moves, where to put them in a prompt, and how to plan a scene that cuts together like a real film.
The camera is your first tool
Two creators can use the same model and get completely different films. The difference is rarely the tool: it is the camera. Where it stands, how close it is and how it moves decide whether a clip feels like cinema or like a demo.
AI video models learned from millions of films, so they understand the vocabulary of directors. If you speak that language in your prompts, you get shots that cut together.
Shot sizes: what each one says
- Extreme wide shot: a tiny figure in a huge place. Says "this world is big" or "she is alone".
- Wide shot: the whole body and the place around it. Shows where we are and who is there.
- Medium shot: from the waist up. The workhorse of dialogue and action.
- Close-up: the face fills the frame. Emotion, decisions, secrets.
- Extreme close-up: an eye, a hand, a key. Tension and detail.
- Over-the-shoulder: we see one character past the other's shoulder. Makes a conversation feel real.
- Insert: an object that matters, a letter, an arrow, a ring. Cheap to make and perfect for hiding cuts.
Vary the size from shot to shot. Two wide shots in a row look like a mistake; wide, then medium, then close-up feels like a film.
Six camera moves AI models understand
Write the move in plain film words. These six work in almost every model.
- Push-in (slow dolly in): we move toward the character. Builds tension or emotion.
- Pull-out (dolly out): we move away. Reveals where the character really is, often a twist.
- Pan: the camera turns left or right from one place. Follows a look or discovers a landscape.
- Orbit: the camera circles the character. Makes a hero moment feel epic; use it rarely.
- Crane up: the camera rises. Ends a scene, or shows the size of an army or a city.
- Tracking shot: the camera moves alongside a walking or running character. Keeps energy in action.
- Push-in
- Pull-out
- Pan
- Orbit
- Crane up
- Tracking shot
- One move per shot. The same scene, six different feelings.
One move per shot
Models get confused when you ask for several moves in one clip, and the result often shakes or warps. Real films do the same: one clear move, or none.
"The camera pans left, then pushes in, then orbits around her and cranes up to the sky."
"Slow push-in on her face as she reads the letter." Then, in the next shot: "Crane up from the house to the snowy forest."
A static shot is also a choice. For dialogue and emotion, "static camera, locked-off" often looks the most professional.
Where to put the camera in your prompt
Most models follow the first words best. Put the shot size first and the camera move near the end, in the same order every time, so your shots match.
- Shot
- Close-up, slow push-in
- Character
- Lyra, red braid, green cloak
- Action
- reads a letter, eyes fill with tears
- drops the letter, turns to the door
- Place
- wooden cabin at night
- same cabin, door half open
- Light
- warm candlelight from the left
- Style
- 35mm film look, shallow depth of field
- Next shot: only these change
- Same order every time. From one shot to the next, change only what must change.
Lens and light words that help
- "35mm" for natural, documentary shots; "85mm" or "telephoto" for soft, flattering close-ups.
- "Shallow depth of field" blurs the background and isolates the character.
- "Handheld" adds urgency to action; "static tripod shot" calms everything down.
- "Low angle" makes a character powerful; "high angle" makes them small or lost.
- "Golden hour", "overcast", "neon at night", "candlelight": name the light, not the mood.
Plan coverage, like on a film set
On a real set, each scene is filmed from several angles so the editor can choose. Do the same with AI: for an important scene, plan a wide shot, a medium shot, close-ups of each character and one or two inserts. Then cut between them.
Coverage costs more generations, but it saves your edit. When a clip fails, you can cut to another angle instead of re-generating the same shot again and again.
Worked example: a letter scene in six shots
- Extreme wide shot, static: a lonely cabin in the snow at night, warm windows.
- Medium shot, slow push-in: Lyra at the table, opening a letter.
- Insert, static: the handwriting on the letter, candlelight flickering.
- Close-up, slow push-in: her eyes fill with tears.
- Over-the-shoulder, static: past her shoulder, the door slowly opens.
- Crane up, outside: from the cabin roof to the dark forest, a silhouette among the trees.
Four different shot sizes, only three moves, and two static shots. That mix is what makes it feel directed.
Ask an AI assistant for a shot list with camera moves
Act as a film director. Here is a scene from my AI series: [describe the scene in 3 to 5 sentences]. Write a shot list of 8 to 12 shots. For each shot give: the shot size, one camera move or "static", the lens (35mm or 85mm), the action in one sentence and the light. Vary the shot sizes, use only one move per shot, and include at least one insert and one over-the-shoulder shot.
Final checklist
- Every shot in your list has a size and at most one camera move.
- Two shots in a row never have the same size.
- The camera words are in the same place in every prompt.
- Important scenes have coverage: wide, medium, close-ups and an insert.
- You used static shots on purpose for dialogue and emotion.
- Name the shot size first and one camera move per shot.
- Vary shot sizes from one shot to the next.
- Plan coverage for important scenes so your edit has choices.
- Pick the right AI video tool for every shot
Text-to-video, image-to-video, start and end frames, lip sync: which method and which model for each kind of shot, how to test them in ten minutes, and how to spend your credits where they count.
Direct the camera: shot sizes and camera moves for AI video — DeepStories Academy