Production playbook
A practical AI video prompt guide
The fastest way to get a usable AI video is to brief one shot clearly. Treat the prompt like a call sheet for a single moment, not a paragraph of mood words or an entire commercial.
Start with a shot, not a story
Long prompts often contain several camera angles, time jumps, actions and locations. That gives the model competing instructions. Instead, generate the sequence as distinct shots: a setup, a detail, a reaction, and a closing frame. You gain control in the edit and can revise one weak beat without throwing away the whole idea.
[subject] [does one action] in [place and time]. [framing and camera movement]. [light and visual treatment]. [pace or physical constraint].
A prompt does not need every field, but it should be specific about the parts that matter. If a product label must remain visible, say so. If the camera must not move, say that too.
The six details that make a shot legible
1. Subject
Name the person, object or scene in concrete terms: “a matte red running shoe” is more useful than “a cool product.”
2. Single action
Choose one visible change: opens, turns, walks, pours, looks up, or comes to rest.
3. Place and time
Give the action a physical setting: kitchen counter at sunrise, wet city street after rain, studio sweep with a soft gray backdrop.
4. Framing and motion
State the starting composition and one camera instruction: close-up, low angle, locked-off wide, slow push-in, or side tracking shot.
5. Light and treatment
Describe the source and quality of light before adding a broad style: window light, hard noon sun, softbox reflection, grainy 16mm texture.
6. Constraint
Use a short constraint when it protects the result: “no cuts,” “label remains facing camera,” “subtle movement only.”
Turn a vague idea into a directable shot
Here is the same idea at three levels of clarity. The last version is not longer for its own sake; each sentence tells the model what the viewer should notice.
Vague Luxury coffee commercial, cinematic and beautiful. Better A ceramic cup of black coffee on a walnut table at sunrise. Slow push-in. Directable A close-up of steam rising from a matte black ceramic coffee cup on a walnut table at sunrise. The camera makes one slow push-in from table height. Warm window light catches the steam; the label on the coffee bag stays sharp in the background. No cuts, quiet morning pace.
Change one variable when a generation misses
When a result is wrong, resist rewriting everything. Keep the successful parts and change the one instruction that failed. That makes the next result easier to diagnose.
- If the composition is wrong, rewrite the framing first: “tight close-up,” “full body in frame,” or “product centered against a clean background.”
- If the motion is chaotic, remove secondary actions and use one camera move. “Slow orbit” is clearer than “orbit, zoom, whip pan and handheld energy.”
- If the mood is off, replace abstract adjectives with a light source, time of day and surface detail.
- If continuity drifts, copy the details that must remain fixed into the next shot rather than assuming the model remembers them.
Before you generate
- Can you point to one clear subject and one visible action?
- Does the shot say where the camera starts and whether it moves?
- Are visual locks—wardrobe, product color, setting, time of day—named when they matter?
- Could you cut this shot next to another one without needing it to explain the whole story?
Put the prompt into a real shot
Use the video studio to choose a model, set the frame, and add camera and lighting direction before you generate.