LARUNDStart free
← All articles

GUIDE · 2026

How to control camera angles in AI video prompts

Most weak AI videos fail on camera, not on content. The subject is fine, but the framing is accidental: a medium shot where you wanted a close-up, a static camera where you wanted a push-in. Models fill gaps with defaults, so you have to say what you want.

This guide gives you a vocabulary and a repeatable formula. It works across models, though each one interprets camera words a little differently, so test your favourite phrases on the model you plan to use.

The four camera decisions

Every shot answers four questions. Shot size: how much of the subject is in frame. Angle: where the camera sits relative to the subject. Lens: how wide or compressed the image looks. Movement: whether and how the camera moves during the clip.

If you answer all four in your prompt, the model has little room to guess. If you answer none, you get an average, centred, mid-distance shot with a slight drift.

Shot size and angle vocabulary

Use standard film terms. Models have seen millions of captions that use them, so they respond well. Choose one shot size and one angle per clip.

TermWhat you getUse it for
Extreme close-upA detail: an eye, a button, a drop of waterTexture and emotion
Close-upFace or object fills the frameReaction, product hero
Medium shotWaist upDialogue, presenters
Wide shotSubject and surroundingsEstablishing a place
Low-angleCamera below, looking upPower, scale, heroic feel
High-angleCamera above, looking downVulnerability, overview
Overhead (top-down)Straight downFood, flat lays, maps
Eye-levelNeutral, at the subject’s heightNatural, documentary feel
Over-the-shoulderBehind one person, looking at anotherConversations

Lenses change the look

Lens choice is shorthand for perspective. A wide lens such as 24mm exaggerates depth and makes spaces feel big. A 35mm lens feels natural and slightly wide, a good default for people in places. An 85mm lens compresses the background and flatters faces. A macro lens shows tiny details with a blurred world around them.

In Larund, you can pin the lens and angle with tokens typed into the prompt, for example &35mm or &low-angle, which makes the intent explicit and easy to reuse across shots. Other tools respond well to the same idea written as plain words, such as “shot on a 35mm lens, low angle”.

Depth of field is a related choice. Ask for a shallow depth of field to separate a subject from the background, or a deep focus when you want everything sharp, such as in landscapes or architecture.

Camera movement: pick one

Movement words are the most powerful and the most abused. Models handle one clear movement well and become unstable when you stack several, because the frame has to stay coherent while everything changes.

Pick one: dolly in, dolly out, pan left or right, tilt up or down, tracking shot following the subject, orbit around the subject, crane up, handheld, or static locked-off. If you want no movement, say “static camera, locked off” explicitly, otherwise the model may add drift.

  • Dolly in: slow push toward the subject builds intensity.
  • Tracking: the camera moves with a walking or driving subject.
  • Orbit: the camera circles the subject, good for products and characters.
  • Handheld: slight shake for a documentary or found-footage feel.
  • Static: best for dialogue, product shots and anything that must stay clean.

A prompt formula you can reuse

Put the camera first, then the subject, action, setting and light. Camera-first prompts tend to be followed more reliably because the framing is established before the details compete for attention.

Example one: “Low-angle, 35mm, slow dolly in. A chef lifts a copper pan, flames rising, warm kitchen light.” Example two: “Overhead, static camera. Hands arrange fruit on a wooden table, soft window light.” Example three: “Close-up, 85mm, shallow depth of field. A woman turns toward a rainy window, neon reflections on her face.”

  1. Choose shot size and angle.
  2. Add the lens or a depth-of-field note.
  3. Add exactly one camera movement, or state that the camera is static.
  4. Describe subject, action and setting in one sentence.
  5. Add light and mood: soft morning light, hard noon sun, neon night.

Lock the angle before you pay for motion

The cheapest place to fix a wrong angle is before the video is rendered. Larund starts each video from approved keyframes: you see the start frame and the end frame, and you only render motion once the framing is right. If the angle is wrong, you change the keyframe prompt, not an expensive clip.

Models differ in how closely they follow camera instructions, so try the same prompt on two or three. Larund lists video models such as Veo 3.1, Kling 3.0, Seedance 2.5 and Runway Gen-4.5 under one credit balance and shows the price before each generation, which makes this kind of comparison practical. For a shot where you need a different angle of the same scene, generate a new keyframe from the same location and cast sheets rather than re-describing everything.

Is Larund the right pick?

A good fit when

  • You want to test one camera prompt across several video models without separate subscriptions.
  • You want to approve the framing as a keyframe before rendering motion.
  • You want lens and angle tokens to pin a shot and reuse it across scenes.

Probably not the best fit when

  • You need exact physical camera paths with numeric control, as in a 3D animation package.
  • You only ever use one model and are happy with the camera controls it already offers.
  • You need to edit footage you already shot with a real camera.

Frequently asked questions

How do I control the camera in an AI video?

Name the shot size, angle, lens and a single camera movement at the start of your prompt, then describe the subject. For example: “Low-angle, 35mm, slow dolly in.”

Why does my AI video ignore camera instructions?

Usually the prompt asks for too many movements, or the model weighs the subject description more heavily. Put the camera first, ask for one movement only and test another model if it still ignores you.

What does a low-angle shot mean in a prompt?

The camera sits below the subject and looks up, which makes the subject feel larger or more powerful. In Larund you can pin it with the token &low-angle.

How do I stop the camera from moving?

Write “static camera, locked off” in the prompt. Without it, some models add drift or a slow push.

What are the &35mm and &low-angle tokens in Larund?

They are lens and camera-angle tokens you type into a prompt to pin the shot, so the framing is explicit and reusable across scenes.

Can I change the angle of an existing shot?

Generate a new keyframe of the same location and character from the new angle, then animate it. This keeps the look consistent better than re-prompting from scratch.

Which AI video model follows camera prompts best?

It varies by model and by shot, and results change as models update. Run the same camera prompt on two or three models and compare; Larund shows the price before each run.

Every top model. One studio.

Start free with Larund Studio: 25+ video and 30+ image models, keyframe films and motion graphics, with the price shown before every generation.

Try Larund free