Camera angles are not style words to sprinkle into an AI prompt. They describe where the viewer is placed relative to a subject, while shot size describes what fits inside the frame and camera movement describes how that viewpoint changes over time. Mixing those controls makes it impossible to know why a generated shot worked—or how to reproduce it.

Build your 12-cell camera-angle board in APOB

This tutorial keeps one scene fixed, changes one camera relationship per cell, and scores the visible geometry. The output is not a glossary. It is a contact sheet, rejection log, and three-shot storyboard a creator can use in the APOB AI Video Generator or another documented workflow. Evidence checked: September 9, 2026.

Separate angle, framing, and movement first

Adobe's guide to camera shots and angles distinguishes distance-based shots, multi-subject framing, angles, and movement. Use that separation in the prompt and in the scorecard.

Viewpoint

Viewpoint is the camera's vertical and horizontal relationship to the subject: eye level, high angle, low angle, overhead, ground level, profile, three-quarter, or over-the-shoulder. Describe a physical relation: “camera 40 centimetres below the subject's eye line, tilted upward.” The visible test is the horizon, eye line, and foreshortening.

Use a simple diagram or written coordinate note beside each prompt. The numbers do not need to describe a real camera rig exactly; they need to preserve the intended relationship. If the output contradicts the diagram, trust the visible geometry and mark the prompt as unfulfilled.

Frame size

Frame size is how much of the subject appears: extreme wide, wide, medium, close-up, or extreme close-up. State boundaries such as “head and upper torso fill the centre third.” A close-up can be high angle or low angle; it does not define viewpoint by itself.

Camera motion

Movement is the change over time: pan, tilt, dolly, truck, crane, orbit, push-in, pull-back, or handheld drift. Test still frames before motion. If the opening geometry is wrong, movement adds another variable and makes diagnosis harder.

When motion is added later, retain an opening-frame checkpoint. A low-angle push-in must begin low and remain coherent; a camera that starts eye-level and drops mid-shot is a different instruction. Label start relation, path, and end relation separately.

Control collision

Flag prompts that contain incompatible instructions: overhead plus eye-level, locked camera plus orbit, extreme close-up plus full-body, or profile plus direct eye contact. Decide which control wins before rendering. The APOB AI Video Prompts page is the natural place to keep the cleaned prompt versions.

Build one scene that can reveal twelve angles

Choose a scene with clear depth, a readable subject, and foreground/background markers. Use a presenter placing a red cup on a wooden counter beside a window, for example. Keep facts neutral and assets owned.

Scene lock

Freeze subject appearance, wardrobe, prop, action, location, time of day, light direction, lens feel, aspect ratio, and output size. Save one base prompt and one reference image. Every cell copies the same text except for the camera-angle clause.

Create a manifest before rendering: base prompt hash, reference-image hash, model surface, visible settings, date, and twelve target clauses. The manifest makes accidental differences obvious and gives a later retest a stable starting point when a model changes.

Twelve cells

Use a balanced board: eye-level, high, low, overhead, ground-level, profile, three-quarter, over-the-shoulder, wide establishing, medium relational, close-up emotional, and extreme close-up detail. The last four alter framing; label them clearly so the board does not pretend every cell is a pure angle.

Render the cells in randomized order if the interface allows it, then restore them to the board order for review. This reduces the temptation to “improve” prompts as the sequence progresses. A changed shared phrase requires a new board version, not a silent continuation.

Negative controls

Add only exclusions that protect the experiment: no extra people, no duplicate cup, no camera movement, no text, and no wardrobe change. Do not use a long negative list that changes interpretation across cells. Keep it identical everywhere.

Contact sheet

Export one representative still from each result and arrange a 4×3 sheet. Label cell ID and prompt clause outside the image, not inside generated content. Keep the original output, selected frame number, and rejection reason. A board without source links becomes a mood board rather than evidence.

Order prompt tokens to protect the geometry

A consistent order does not guarantee obedience, but it makes tests comparable. Use: subject anchor → camera relation → frame boundary → action → environment → light → style tail. Change only the target clause.

Subject anchor

Start with the same concrete subject and prop: “One presenter in a dark green shirt holds one red ceramic cup.” Countable nouns reduce ambiguity. Keep identity reference and action wording unchanged.

Camera relation

Write the viewpoint as a physical instruction before aesthetic language: “camera positioned above the presenter, angled down 35 degrees.” Avoid stacking synonyms such as bird's-eye, top-down, overhead, aerial, and drone in one cell; the pile obscures which token mattered.

Keep the relation clause short enough to compare by eye. Save one alternate clause only after a failure, and label it as a repair run. The original failed wording belongs in the rejection log because it teaches future prompt writers which formulation was ambiguous on this surface.

Frame boundary

State what must remain visible and where it sits: “medium shot, head to waist, both hands and cup visible, counter edge in lower frame.” This protects composition when the camera relation changes. For image-led experiments, the APOB Image to Video AI workflow can preserve a chosen starting composition while motion is tested separately.

Write an acceptable crop range rather than demanding pixel-perfect placement. The test asks whether the intended camera relationship and narrative information survive, not whether a generative output matches a mechanical layout. Reject only when a missing boundary removes required evidence or changes the shot job.

Style tail

Put colour, lighting, texture, and genre after geometry. Use one restrained style clause for all twelve cells. If “dramatic cinematic lighting” changes the apparent angle through heavy shadow or lens distortion, remove it and rerun.

Example token order:

[subject] + [camera relation] + [frame boundary] + [action] + [setting] + [light] + [style] + [fixed negatives]

Version the prompt. Do not quietly repair only the failed cell; duplicate the board as a new run if the shared wording changes.

Score what the image actually shows

Never pass a shot because its filename says “low angle.” Review observable geometry. Score zero for absent, one for ambiguous, two for mostly correct with a repairable issue, and three for clear and usable.

Horizon

Mark the horizon or dominant horizontal reference. Its height and tilt help reveal viewpoint, but interior scenes may not have a literal horizon. Use counter edges, door frames, or background planes as substitutes and note the limitation.

Draw the reference line on a duplicate still, never on the evidence original. Store the unmarked frame, annotated frame, and frame number together. The annotation should explain the judgment rather than hide an uncertain angle behind a coloured arrow.

Eye line

Compare the camera with the subject's eyes and gaze. In a high-angle shot the viewer should plausibly look down; in a low-angle shot, up. If the face is merely tilted while the camera remains eye-level, reject the angle claim.

Foreshortening

Inspect the relative scale of near and far features. A low, close viewpoint may enlarge the foreground hand or chin; an overhead view reveals top surfaces. Distortion is evidence only when it matches the intended camera relation and remains usable.

Compare foreshortening with the reference object as well as the face. A cup, counter edge, or hand can reveal depth more reliably than an isolated portrait. If style changes object proportions, mark both geometry confidence and prop integrity so the reviewer does not confuse distortion with a successful angle.

Continuity

Compare face, wardrobe, prop, hand position, light direction, and screen direction across winning cells. The most dramatic individual frames may fail as storyboard camera shots if they cannot cut together.

Add a continuity score beside the geometry score. A cell can pass its angle and fail the storyboard. This two-axis result prevents the team from selecting a spectacular low angle whose subject, prop, or screen direction no longer belongs to the scene.

Keep a rejection log with cell, target, observed geometry, conflicting cue, continuity issue, and next action. This turns failed AI video prompts into reusable production knowledge.

Turn three winners into storyboard coverage

Choose three frames by narrative job, not visual novelty. The BBC Academy's Five Essential Shots shows how changes in size and angle help shots edit together while warning against crossing the line. Adapt that logic to your generated scene.

Establishing frame

Select a wide or high-context frame that shows subject, counter, window, and action space. Confirm geography and screen direction. The frame should answer where the scene happens before it asks the viewer to inspect emotion or detail.

Annotate the intended cut direction from the establishing frame into the relational frame. If the second shot flips the subject across the line of action, choose another winning cell or rerun it. Strong individual camera shot angles still need to form a coherent sequence.

Relational frame

Choose a medium, profile, or over-the-shoulder view that explains the subject's relationship to the cup and surface. Both action and relevant object must be legible. This shot carries the practical beat.

Emotional frame

Choose a close-up whose eye line and expression match the story moment. Do not sacrifice identity continuity for intensity. If the prompt changes the person's age, wardrobe, or lighting, the frame is not a winner.

Reshoot trigger

Write a trigger for each shot: wrong screen direction, missing hand contact, uncertain angle, identity drift, duplicated prop, or uneditable transition. Assign the smallest repair—new angle cell, new frame boundary, or motion test—without rebuilding the whole board.

The result is a production artifact: twelve labelled experiments, a geometry rubric, rejected-output evidence, and three compatible frames. That is more useful than memorizing camera angle prompts because it shows what this scene and this workflow actually produced.

Archive the base prompt, twelve clauses, reference image, visible model settings, original outputs, selected frame numbers, contact sheet, annotated geometry, continuity scores, and storyboard order. Keep the prompt worksheet, test matrix, generation receipt, failure ledger, acceptance checklist, verdict record, and review overlay template in the same folder.

Add the run date and a recheck trigger. When a model or prompt interpreter changes, rerun the same board before replacing the accepted frames; do not compare a new workflow with an undocumented memory of the old one.

For the next scene, keep the control families and scoring method but rebuild the observable cues. A kitchen counter reveals depth differently from an outdoor horizon or a talking-head setup. The method stays reusable because it measures what each scene can actually show.

Keep every decision dated.

Sources

• Adobe — Different Types of Shots and Camera Angles

• BBC Academy — Five Essential Shots

• APOB AI — AI Video Prompts

• APOB AI — AI Video Generator

• APOB AI — Image to Video AI