Product video prompts that keep the product the same
A good product video prompt does not describe the product. The reference images already show it; the prompt places it in a scene, gives it one clear motion and tells the camera what to do. Below are full prompts for hero shots, in-use shots and lifestyle scenes, each with a note on why it holds up and what to change when it does not.
What makes product prompts work with references
Refer to the product by its role, such as "the bottle", or by its saved name, and stop there. A second description of its color or label competes with the images, and the model often splits the difference.
Give each clip one motion: a turn, a pour, a lift. Two actions in six seconds usually blur into neither. Name the camera move, the light source and the end frame, because a defined ending ("ends with the front label facing the camera") lands on an angle your references actually show.
Plan around small text. Fine print is re-drawn by every current model, so keep labels large and frontal, or let them fall out of focus.
Length, shape and sound are given in brackets before each prompt; they are settings, not prompt text.
Hero and turntable shots
Prompt (6 s, 16:9): The bottle stands on a matte white plinth in the center of the frame. It rotates slowly clockwise through a half turn and stops with the front label facing the camera. The camera holds at eye level and pushes in slightly over the last two seconds. Soft overhead light through a large diffuser, a faint reflection on the plinth, plain light-grey background.
Why it works: a half turn ends on the front view, so the model only has to invent the back for a moment. If the label warps mid-turn, shorten the rotation to a quarter turn.
Prompt (5 s, 16:9): Macro close-up of the watch resting on dark slate. The camera slides left to right across the dial with a shallow depth of field, focus moving from the crown to the logo. One hard light from the upper right draws a crisp highlight along the edge of the case. Bare tabletop, still air.
Why it works: a slow slide shows texture and finish, which is what buyers look for. Include a sharp dial close-up in the reference set, or the logo will be approximated.
Prompt (8 s, 16:9): The headphones hang in midair against a deep blue gradient. They turn a quarter turn to the left while the camera orbits a quarter turn to the right, ending on a three-quarter view. A cool white rim light outlines the ear cups and a soft fill keeps the front readable.
Why it works: two small opposite motions look like one large one, and neither strains the model. If the headband bends, drop the product rotation and keep only the orbit.
Product in use
Prompt (5 s, 16:9): A hand picks up the mug from a wooden kitchen counter, lifts it toward the camera and carries it out of frame to the right. Steam rises from the coffee inside. Morning window light from the left, warm tones, the kitchen soft and blurred behind. Static camera at counter height.
Why it works: the hand enters, does one thing and leaves. A full grip around the body is easier for the model than fingers threading through a thin handle.
Prompt (7 s, 9:16): The backpack sits on a park bench. A person in a plain grey jacket swings it onto one shoulder and walks away down a city sidewalk. The camera follows at waist height for two seconds, then stops and lets them walk out of focus. Overcast daylight, soft shadows.
Why it works: a plainly dressed person never competes with the product. If straps or zips change shape, add a back-view photo to the references.
Lifestyle and context scenes
Prompt (8 s, 16:9, sound on): The candle burns on a windowsill at dusk, a city skyline out of focus behind it. The flame flickers gently while the camera pushes in from a medium shot to a close-up of the jar. Blue light outside, warm glow on the glass. Ambient sound: distant traffic and quiet room tone.
Why it works: the cool-warm contrast gives mood without props, and the push-in finishes on the product. With sound off, leave out the sound line.
Prompt (5 s, 16:9): The running shoe lands on wet asphalt in slow motion and splashes a shallow puddle, seen from ground level. The camera tracks alongside at the same speed. Early morning, low sun behind the runner, backlit droplets.
Why it works: ground level shows the side and sole, the parts shoppers compare. That only holds if a sole photo is in the reference set.
Using these prompts in RefVid and elsewhere
In RefVid a product is saved once with 1 to 5 photos and attached to every scene, so prompts like these only need to carry the scene. With one product, the role word is enough. With two or three products in one shot, write each saved name; RefVid swaps the names for Kling 3.0 Omni's @Element tags. Clips run 3 to 15 seconds, which rarely needs to be longer than 8 for a product shot.
Product photos can be imported from a shop link, and a missing angle can be drawn from one photo. Describing what you want also works better than listing exclusions: "a bare white tabletop" is clearer than "no props".
On Veo 3.1, as of October 2026, according to Google's Veo docs, clips with reference images are 8 seconds and take up to three reference images, so keep hero shots to one motion and pick your three best angles.
Frequently asked questions
- Should I describe the product's color and logo in the prompt?
- No. The reference images carry that information. A written description competes with the images and can shift the color. Describe the scene, motion, camera and light instead.
- How long should an AI product clip be?
- Five to eight seconds with one motion is a good default. Longer clips give the model more time to drift, and most ads cut every few seconds anyway.
- Why does the text on my product label change in AI video?
- Small printed text is re-drawn by current video models. Keep labels large and facing the camera, add a sharp label close-up to the references, or let the text fall out of focus.
- Can I show two products in one AI video shot?
- Yes. Give each product its own references and name each one in the prompt. In RefVid, up to 3 saved subjects can share one shot, each named by its saved name.