How to Write Veo Prompts for E-commerce Product Videos

The first time a production team moves from image-generation prompting to Google Veo, the pattern is almost always the same. Someone types “product shot on a clean white background” — the exact phrasing that worked flawlessly in Midjourney or DALL-E — and gets back a clip where the item morphs into a different shape by frame 30, the camera drifts sideways for no reason, and the background shifts from white to beige to gray across the duration. The team burns through credits, blames the model, and eventually realizes the problem was never the vocabulary.

Veo prompts are temporal documents, not descriptions. They encode subject, motion, camera, and environment as an ordered sequence that the model resolves across every frame. E-commerce product videos succeed or fail on that structure, not on how many adjectives you stack in front of the product name. A prompt that reads like a static image brief will produce a video that falls apart by the third second.

Why Veo Prompts Behave Differently from Image Prompts

Veo interprets a prompt as a shot sequence rather than a still frame. Where an image model only needs to resolve a single composition, Veo has to maintain motion continuity, physical plausibility, and camera behavior across dozens of frames. Every token you write gets distributed across the entire temporal span of the clip, which means vague language doesn’t just produce a slightly off image — it produces an object that drifts, morphs, or changes material halfway through the shot.

This is the core mismatch that catches most teams. Image prompting mental models carry over badly because they treat the prompt as a description of a moment. Veo treats it as a description of a process. When you say “a sneaker on a pedestal,” the model has to decide what the sneaker does over time, how the camera behaves, and whether the environment stays static. Leave any of that unspecified, and the model fills the gap with whatever its training data suggests — which is rarely what you wanted.

Veo 2 generates native clips up to 8 seconds long in resolutions up to 4K. That duration is long enough that prompt errors compound across frames. A small ambiguity in the first second becomes a visible morph by the fifth. The model isn’t failing randomly; it’s resolving your underspecified prompt in the most statistically probable way, and statistical probability rarely matches art direction.

The practical takeaway: treat every Veo prompt as a miniature screenplay. Subject first, then action, then camera, then environment. If you skip any of those elements, the model invents them.

The Anatomy of a Structured Veo Prompt

A reliable Veo prompt for product video breaks down into five components: subject, action or motion, camera movement, lighting, and environment or style. The ordering matters more than most people expect. Veo weighs early tokens more heavily, so the subject and its primary motion need to appear in the first few words. Bury the action at the end of a 60-word prompt, and the model may simply ignore it.

Keep prompts to roughly 30–60 words. Longer prompts risk the model dropping mid-prompt descriptors, especially ones that appear after the camera instructions. This is a known tradeoff: brevity gives the model less to work with, but it also prevents the trailing half of your prompt from being silently discarded. A tight 40-word prompt that sequences the five components in order will outperform an 80-word prompt that lists everything at once.

The table above shows the difference between a vague line and a structured one. Notice that each structured example specifies exactly one thing per element. That’s deliberate. When you give Veo two competing motions or two conflicting lighting directions, it resolves the conflict by averaging them, which produces a clip where neither reads clearly.

Consider a tightly structured food prompt as a worked example. A round raw pizza dough prompt that specifies the dough’s shape, its motion, and the camera behavior in sequence will hold together across the full clip. The same subject described as “pizza dough on a counter” will drift and deform because the model has no instruction about what the dough should do over time.

The tradeoff between brevity and control is real. Shorter prompts give the model freedom to produce surprising results, which is sometimes desirable for mood pieces. But for e-commerce product video, where the client needs the product to look exactly like the product, control wins. Every word you add that constrains motion or environment reduces the chance of an off-brief frame.

Writing Veo Prompts for E-commerce Product Showcases

E-commerce scenarios cluster into a few repeatable formats: hero product shots, packaging close-ups, lifestyle demonstrations, and short commercial-style spots. Each format needs a different motion script, but the underlying prompt structure stays the same.

For a hero shot, the motion should be simple and legible. The product enters frame, rotates once, and settles into a static position. That’s one action, clearly sequenced. A footwear example works well here — a shoe landing on a reflective black surface prompt that specifies the drop, the impact, and the resulting stillness gives the model a clear arc to resolve. Contrast that with “a shoe on a surface,” which leaves the model to decide everything about how the shoe arrives.

Packaging close-ups need even tighter motion control. The camera should move less, and the product should barely move at all. A slow dolly toward the label, with the product stationary, reads as premium. A rotating bottle with a moving camera reads as chaotic. Specify the camera movement explicitly and keep the product motion minimal.

Lifestyle demonstrations are where the motion script gets more complex. The product needs to be used, which means you have to describe the action in enough detail that the model can resolve it frame by frame. “A person applies the moisturizer to their forearm” is a start, but you’ll get better results with “a person pumps the moisturizer onto their hand, then smooths it onto their forearm in two slow strokes.” The added specificity about the pump and the stroke count gives the model a concrete sequence to follow.

Commercial-style spots are the hardest because they require multiple elements to stay coherent. A KITKAT commercial-style prompt that sequences the product reveal, the break, and the camera movement will hold together better than one that lists three separate actions without ordering them. The key is to script the motion as a single continuous flow rather than three disconnected events.

A single hero-shot prompt typically needs 3–5 generation attempts before the motion reads cleanly. This isn’t a sign of a bad prompt; it’s the normal iteration cost of getting temporal coherence right. Teams that budget for multiple attempts per shot are the ones that end up with usable footage.

Aspect ratio matters more than most people think. Square formats work for social feeds, 16:9 works for product pages and video ads, and 9:16 works for short-form vertical platforms. Specify the aspect ratio in the prompt or the generation settings, because the model will otherwise default to whatever it was trained on most recently. Changing the aspect ratio after generation means regenerating — there’s no reliable way to crop a Veo clip without losing composition.

Iterating and Refining Veo Prompts in a Production Workflow

Prompt writing for Veo is an iterative loop: draft, generate, inspect the failure mode, revise one variable, repeat. The teams that treat it as a one-shot writing task burn through credits and end up with nothing usable. The teams that treat it as a debugging workflow get consistent results.

The common failure patterns map to specific prompt elements. Object morphing traces back to an underspecified subject — the model didn’t have enough detail about the product’s shape or material to hold it stable across frames. Camera jitter traces back to vague camera instructions — “camera moves” gives the model no direction, so it produces micro-movements that read as handheld shake. Background bleed traces back to an unspecified environment — the model fills the void with whatever its training data suggests, and that suggestion changes between frames.

When a generation fails, change exactly one variable. If you revise the subject description and the camera movement in the same iteration, you won’t know which change fixed the problem. This is the same discipline as debugging code: isolate the variable, test, observe, repeat.

Teams that batch-test several prompts against the same brief and log which phrasing wins build a de facto prompt library. The logging step is what most people skip, and it’s the difference between a team that improves over time and a team that starts from zero with every new product.

The friction of rebuilding prompts from scratch for every new product is real. A validated prompt skeleton — the same subject-motion-camera-lighting-environment structure with the product-specific details swapped in — cuts per-clip iteration time by roughly half. Teams that log and reuse a skeleton don’t have to rediscover the motion script that worked last quarter.

This is where a centralized prompt library earns its keep. VideosPrompt aggregates community-validated prompts across product showcases, commercials, and food photography, which means a team can start from a proven structure instead of drafting from a blank field. The marginal prompt is easier to debug when only one variable changes between iterations, and starting from a validated base reduces the number of variables you have to guess at.

A reusable commercial concept is worth more than a one-off prompt. Consider a restaurant menu reveal in 3D cinematic style — the structure of that prompt, with its sequenced reveal and camera movement, can be adapted to any product that needs a dramatic introduction. The product changes, the skeleton stays.

The teams that succeed with Veo at scale are the ones that treat prompts as assets rather than disposable text. They version them, log the results, and reuse the structures that work. The teams that treat every prompt as a fresh writing task are perpetually stuck at the bottom of the learning curve, paying for the same mistakes repeatedly.

FAQ

How long should a Veo prompt be for a product video?

Aim for 30–60 words. Shorter prompts risk underspecifying the motion or environment, which leads to the model inventing details. Longer prompts risk the model dropping trailing descriptors, especially ones after the camera instructions. The sweet spot is a tight sequence of subject, motion, camera, lighting, and environment that fits in roughly 40 words.

What is the best way to describe camera movement in a Veo prompt?

Specify one movement, its direction, and its speed. “Slow dolly in toward the label” works better than “camera moves closer.” Avoid stacking multiple camera movements in a single shot — the model will average them into a jittery mess. If you need a complex camera move, break it into a sequence and describe each segment separately.

Why does Veo change the background or object between frames?

The model is resolving your prompt across the full temporal span, and any element you leave unspecified gets filled in statistically. If the background isn’t described, the model generates a different plausible background for different frame ranges. Specify the environment explicitly — surface, color, lighting — and the object will hold stable because the model has fewer gaps to fill.

Can the same Veo prompt be reused across different products?

Yes, if you keep the structure and swap the product-specific details. The subject line changes, but the motion script, camera movement, and environment stay the same. This is why a validated prompt skeleton is valuable — it lets you reuse the parts that worked and only debug the parts that changed.

Which settings matter most when generating video for e-commerce storefronts?

Aspect ratio and resolution matter most. Square for social feeds, 16:9 for product pages, 9:16 for vertical platforms. Resolution matters because e-commerce footage often gets cropped or zoomed in post-production, and a 4K source gives you room to work. Duration matters less — most product shots only need 4–8 seconds of usable footage.