Video prompt composition is the language you use to tell an AI model how to frame a shot, wide or close, where the subject sits, how the eye should travel, and it is the single biggest lever on whether a generated shot looks intentional or accidental. Wide, medium, close, rule of thirds: these are not jargon. They are instructions a model can act on. This guide covers the composition terms that work, how to apply them per shot, and how an agentic Director locks framing into the plan before a single frame is generated.
Why composition is a prompt, not an afterthought
A model generates what you describe. Leave composition out of your prompt and the model picks for you, usually a safe, centered, middle-distance frame that reads as generic. Specify composition and you reclaim the most powerful storytelling tool in cinematography: where to put the camera and what to put in the frame.
Composition does three jobs at once. It directs attention, telling the viewer what matters. It sets scale, telling the viewer how big the world is. And it carries emotion, a tight close-up feels intimate or tense, a wide shot feels lonely or grand. When you skip composition language, you forfeit all three and hope the model guesses right. When you include it, every shot becomes a deliberate choice, this is the foundation of good prompting generally, and it pairs naturally with the anatomy of a great AI video prompt.
The composition vocabulary models understand
Models respond best to concrete, conventional film terms, the same ones a cinematographer would use on set. Vague mood words like "cinematic" or "beautiful" get ignored; spatial instructions get followed.
Shot size: wide, medium, close
Shot size is the first decision and the most important.
- Wide / establishing shot: The subject is small in a large frame. Use it to set place and scale. "Wide shot of a lone figure crossing an empty parking lot at dawn."
- Medium shot: Subject from roughly the waist up. The workhorse for dialogue and demonstration. "Medium shot of a barista steaming milk, shallow depth of field."
- Close-up: A face, a hand, an object filling the frame. Use it for emotion and detail. "Close-up of weathered hands turning a brass key."
- Extreme close-up: A single detail. "Extreme close-up of a water droplet hitting a leaf."
Naming the shot size is the fastest way to move a prompt from generic to directed.
Where the subject sits: rule of thirds and balance
The rule of thirds places the subject a third of the way into the frame rather than dead center, which reads as more dynamic and leaves room for the eye to travel. Tell the model: "subject positioned on the left third, looking toward open space on the right." You can also call for symmetry ("centered, symmetrical composition") when you want stillness or formality. The point is to decide where the subject sits rather than defaulting to the middle.
Camera angle and height
Angle changes meaning. A low angle looking up makes a subject powerful; a high angle looking down makes it vulnerable. Eye level is neutral and trustworthy. Specify it: "low-angle shot looking up at the speaker against a bright sky."
Depth and lens feel
Depth of field controls focus. "Shallow depth of field, background softly blurred" isolates a subject; "deep focus, everything sharp" includes the whole scene. Lens language, wide-angle for grandeur and slight distortion, telephoto for compression, shapes how space reads.
How to set framing per shot in the plan
Knowing the vocabulary is half the battle. The other half is applying it consistently across an entire film, shot by shot, which is exactly what an agentic Director's PLAN stage is for.
When you describe a film, the Director boards each shot before anything is generated, assigning it a role, a camera move, and a prompt, and the prompt is where composition lives, that means framing is not a global setting smeared across the film; it is a per-shot decision you can see and adjust on the board. You can make the hook a tight, tense close-up, the establishing shot a sweeping wide, and the payoff a medium that lets a reaction land, all in one plan. The role-based approach is detailed in prompting establishing shots and hooks.
Composition also works hand in hand with movement. Framing decides where the shot starts and ends; motion decides how it gets there. A push-in turns a medium into a close-up over time, so the two should be planned together, the companion piece on prompting motion and action in video covers the movement half. Because planning and scoring are free and only generation costs credits, you can refine the composition of every shot on the board before committing a single credit. You can see how shot roles and framing map into the editable JSON plan in the docs.
Composition patterns you can reuse
A few reliable patterns, ready to adapt by swapping in your subject:
- Intimate testimonial: "Medium close-up, subject on the right third, shallow depth of field, soft window light from the left, eye level."
- Product hero: "Centered, symmetrical composition, single product on a reflective surface, deep focus, hard key light from above."
- Epic establishing: "Extreme wide shot, subject tiny in lower third, vast landscape filling the frame, golden-hour backlight."
- Tension beat: "Low-angle close-up, subject off-center on the left, dark negative space on the right, cool tones."
Notice the consistent order: shot size, subject placement, depth, light, angle. That sequence is dependable across models because it moves from the broadest decision to the finest. Building a small library of patterns like these is the fastest way to make every film feel composed; our text-to-video prompting guide goes deeper on turning patterns into repeatable prompts.
How the quality gate protects your framing
Good composition in the prompt only matters if the generated shot honors it. This is where the quality gate earns its place. Every generated shot is scored on concrete checks, including prompt match, which evaluates whether the shot you got is the shot you described. If you called for a low-angle close-up and the model returned a flat medium, that mismatch is caught and the shot is selectively regenerated rather than accepted. Your composition decisions are enforced, not merely suggested.
That enforcement is what makes per-shot framing trustworthy at film length. You can compose twenty distinct shots knowing each one is checked against its prompt, so the deliberate frame you wrote is the frame that survives into the cut. For the full picture of how directed prompting fits together, the AI video hub collects the craft in one place.
Frequently asked questions
How do I prompt framing in AI video?
Use concrete film terms in this order: shot size (wide, medium, close), where the subject sits (centered, or on a third), depth of field, light, and camera angle. For example, "medium close-up, subject on the right third, shallow depth of field, soft light from the left" gives a model clear instructions, whereas vague words like "cinematic" get ignored.
What composition terms should I use?
Lean on the vocabulary cinematographers use: wide or establishing shot, medium shot, close-up, and extreme close-up for size; rule of thirds and symmetry for placement; low-angle, high-angle, and eye level for camera height; and shallow or deep depth of field for focus. These conventional terms are the ones models reliably act on.
Can I set framing per shot?
Yes. In the Director's PLAN stage, every shot is boarded with its own prompt, camera move, and role before anything is generated, so composition is a per-shot decision you can see and adjust on the board, you can make the hook a tight close-up and the establishing shot a sweeping wide within a single plan.
What if the generated shot ignores my framing?
The prompt-match check in the quality gate scores whether the shot you got matches the shot you described, including its composition. If the framing drifts from your prompt, the shot is selectively regenerated rather than accepted into the cut, so your composition decisions are enforced rather than merely suggested.
Compose every shot on purpose
Video prompt composition turns framing from a roll of the dice into a deliberate instruction, and an agentic Director locks that instruction into the plan, per shot, before anything renders. Decide where the camera goes, where the subject sits, and how the eye travels, then let the quality gate make sure the model honors it. Start free with no credit card and direct your next film frame by frame.