Product Demo Video Prompts

TL;DR: A product demo prompt has an honesty constraint most templates skip entirely: a generated video of a product that doesn't match the real thing is misrepresentation, not a style choice. The safe structure is three distinct shot types, a hero shot, a detail pass and an in-use scene, each doing one job, built from your product's actual appearance rather than an invented one.

Most "product demo video prompt" content treats the job as pure aesthetics, better light, a nicer background, a more cinematic camera move. That's real, and this post covers it. But it skips past the actual risk in this specific category: a demo video that shows features, a UI, or an appearance your product doesn't have isn't a creative choice, it's a factual claim about a real thing, made to a customer who's deciding whether to buy it. The structure below is built around that constraint, not around it despite it. None of what follows argues against making a beautiful demo. It argues that beauty and accuracy aren't in tension here, and treating them as though they were is how a well-lit, well-composed clip ends up making a claim about the product that isn't true.

The line a product demo prompt shouldn't cross

A demo video is a claim. It tells a viewer "this is what the thing looks like" or "this is what it does." Generated video makes that claim exactly as easily whether it's true or not, and a viewer generally has no way to tell the difference from the clip alone. That's the entire reason this category needs a stricter rule than a travel shot or a mood piece: nobody buys a mountain because a landscape video looked appealing, but people do buy products because a demo made a specific claim about them.

Three uses stay clearly on the right side of that line: a concept or mood piece labeled plainly as a concept, not a real capture; B-roll and lifestyle footage that surrounds real footage of the actual product rather than replacing it; and an animated or visibly abstract explainer of how a mechanism works, rather than a photorealistic render standing in for the real object. The three shot types below assume you're demoing something real, described accurately, not invented.

The hero shot

The hero shot is a clean, controlled presentation: one object, a simple background, even light, minimal or no other action in frame. It's the shot doing the least interpretive work and, for that reason, the one where accuracy is easiest to hold onto, describe the object as it actually is and let the camera do the presenting.

Google's own Veo 3.1 documentation lists camera positioning as an optional prose element with its own vocabulary, "aerial view, eye-level, top-down shot, dolly shot", none of which is a request field. For a hero shot specifically, a slow orbit or a static composition read as intentional; a shaky or rapidly moving camera undercuts the "clean presentation" the shot is for.

{
  "prompt": "A matte ceramic pour-over coffee dripper sitting on a plain light-grey studio surface, soft even studio lighting from above and slightly to the side, no visible shadow harshness, camera holding a slow, smooth orbit around the object at eye level, nothing else in frame",
  "config": {
    "aspectRatio": "16:9",
    "resolution": "1080p",
    "durationSeconds": "8"
  }
}

That's a placeholder object described plainly and accurately. Your version replaces the dripper's description with your own product's real shape, material and color, taken from what it actually looks like, not from what would look most impressive.

The detail pass

The detail pass moves in close: material, texture, a specific mechanism or edge. Google's Veo documentation lists "focus and lens effects" as its own optional prose element, giving "shallow focus, deep focus, soft focus, macro lens, and wide-angle lens" as example terms, again prose, not a request field, on any of the three vendors checked for this post. "Macro lens" as a phrase is exactly the right register for this shot type.

{
  "prompt": "Extreme close-up, macro lens, on the textured matte glaze of a ceramic surface, camera holding a slow, minimal drift left to right, single soft light source raking across the surface to reveal fine texture, shallow focus with the immediate foreground sharp and the edges softening",
  "settings": {
    "aspect_ratio": "1:1",
    "duration": 5,
    "audio": "off",
    "multi_shot": false
  }
}

1:1 shows up here deliberately: Kling 3.0 Omni is the only one of the three vendors checked in this post whose settings.aspect_ratio includes a genuine square option alongside 16:9 and 9:16, useful for a detail shot destined for a square product-grid placement rather than a full-width hero position.

The in-use scene

This is where the honesty constraint matters most, because it's the shot most likely to claim a feature or interaction rather than just an appearance. A hero shot or detail pass can misrepresent looks; an in-use scene can misrepresent function, a stronger claim.

{
  "promptText": "A person's hands pouring hot water in a slow circular motion over coffee grounds inside a ceramic dripper, seen from above at the kitchen counter, steam rising, warm morning light through a nearby window, everyday kitchen background softly out of focus",
  "ratio": "1280:720",
  "duration": 8,
  "seed": 305217
}

Runway Gen-4.5's request body for this stays to promptText, ratio, duration, seed, contentModeration and outputFormat, confirmed directly against its published schema, no audio field on generation at all. For a lifestyle or in-use scene where ambient kitchen sound matters, that gets added after generation on Runway specifically, not requested in the prompt the way it can be on Veo or Kling.

A practical check before writing an in-use prompt: read the sentence back and ask what it implies the product does, separately from how it looks. "Pouring water over coffee grounds" implies exactly one thing, that the dripper holds water and grounds while someone pours, which is true of the placeholder object in this example. A sentence that instead implied the same dripper "keeps coffee hot for six hours" or "brews automatically at a button press" would be describing a different product, one with a feature this one doesn't have, dressed up as an in-use shot of this one. The hero shot and detail pass fail this test rarely, since they mostly describe appearance; the in-use scene is where an unearned feature claim slips in easiest, because action reads as capability even when the prompt never states it outright.

Grounding a demo in what the product actually looks like

Description alone asks a model to imagine your product from scratch, which is exactly where accuracy starts to slip: a detail invented to sound impressive is still invented. All three vendors checked for this post accept real reference images instead of description alone, and using them is the single most direct way to keep a demo honest rather than merely well-written.

Veo 3.1's request schema documents a referenceImages field, described on the same page as "Up to three images to be used as style and content references," and separately documents an image field for an initial frame to animate plus a lastFrame field for interpolating toward a second image. Kling's image-to-video endpoint, /image-to-video/kling-3.0, takes a contents[] array typed with first_frame and last_frame image entries rather than a plain prompt-only call. Runway's image-to-video branch takes a promptImage field alongside the same ratio, duration and seed parameters as its text-only branch.

None of that guarantees a pixel-accurate result. It does mean the model starts from your product's real photograph rather than from whatever a text description happens to suggest, which matters most for exactly the details a written prompt is worst at controlling, precise proportions, a specific logo placement, an exact color match. For a hero shot or detail pass where getting the object right is the entire point, feeding in a real photo of the actual product is worth the extra step over prompting from description alone.

{
  "prompt": "The product shown in the reference image, photographed in a clean studio setting, soft even lighting, camera holding a slow orbit at eye level, no other objects in frame",
  "image": "<your own reference photo of the real product>",
  "config": {
    "aspectRatio": "16:9",
    "resolution": "1080p"
  }
}

That template is deliberately generic in its prose and specific in its image field: the description carries only the setting and camera direction, and the actual appearance of the product comes from the reference photo, not from adjectives trying to describe it into existence.

When the real product isn't ready to show yet

Three structures stay honest when a literal, accurate demo genuinely isn't achievable:

E-commerce crossover and where this connects

Most product demos for e-commerce end up needing both a still-image set and a video, and the honesty constraint in this post applies to the stills too, a listing photo that shows a feature the product doesn't have is the same problem in a different medium. If the still-image side of the same job, listing photos rather than video, is the more immediate need, our free product photography prompt generator covers that ground directly. For B-roll specifically as its own technique, independent of the product-demo honesty question, generating B-roll with an AI prompt library goes deeper than this post does. If camera-movement vocabulary for the hero shot needs a fuller reference than the handful of terms used above, camera movement vocabulary for AI video has the complete list, and pacing and timing in video prompts covers how to describe the speed of a shot once more than one is stitched together into a full demo sequence, hero shot to detail pass to in-use scene, rather than generated as three unrelated clips. If the model choice itself is still open, Veo 3 vs Sora vs Kling is the place to start.

One closing note on scope: every field, enum and quote above belongs to Google, Kling AI or Runway, read from their own documentation on September 3, 2026. We build the prompt and the library it lives in. We don't generate the video ourselves, and the honesty constraint in this post applies whether or not any particular tool makes fabrication easy: what a product actually looks like and does is a fact about the product, not a setting on a model.