To prompt Kling 3.0 for ecommerce product shots, start from a clean photo of the real product as the start frame, then write a short prompt that names one camera move, one light change and one small motion, and nothing about how the product looks. Kling 3.0 generates clips of up to 15 seconds from text or images, so the product's shape, label and colour should come from your image while the prompt only directs what happens around it.
Key Takeaways
- Lock the product with an image. Image-to-video with a start frame (and optionally an end frame) keeps packaging, logos and proportions far closer to the real thing than text-to-video.
- Write prompts like a shot list, not a description. One camera move, one lighting cue, one action per shot, in the order the viewer sees them.
- Short clips are easier to control. Generate 5 second product beats and cut them together in the edit instead of asking for one 15 second clip that has to get everything right.
If you would rather hand the whole production over, Adsome makes performance video ads from your real product.
What is Kling 3.0?
Kling 3.0 is the third generation of Kuaishou's Kling AI models, announced on 5 February 2026 as a lineup of Video 3.0, Video 3.0 Omni, Image 3.0 and Image 3.0 Omni. According to the launch announcement, Video 3.0 generates clips of up to 15 seconds, produces native audio with speech in English, Chinese, Japanese, Korean and Spanish, and the Omni model adds a multi-shot storyboard where you set the duration, shot size, perspective and camera movement for each shot.
A product shot is a short clip whose only job is to show one product clearly, such as a pack rotating on a plinth or a trainer landing on wet tarmac. If you are new to the model, our Kling 3.0 Master tier review and tutorial covers the interface and tiers before you start prompting.
Why text-only prompts fail on real products
A text-to-video prompt asks the model to invent your product from words. You get a plausible bottle, but not your bottle: the cap drifts, the label turns into near-letters and the colour lands a shade off.
Image-to-video fixes most of this. A start frame is the image the clip begins from, so every pixel of the product in the first frame is yours. Some providers also accept an end frame, the image the clip must finish on. In Leonardo's Kling 3.0 API documentation, for example, an end frame is only accepted when a start frame is also given, and when you do not set width and height the output takes its dimensions from the start frame. That last point matters: the shape of your source image decides whether you get a 9:16 Reel or a 16:9 YouTube cut.
Which Kling 3.0 input mode fits a product shot
Each input mode trades control for freedom. The table compares them on the criteria that decide a product shot.
| Input mode | What you supply | Product fidelity | Clip length | Best use |
|---|---|---|---|---|
| Text to video | Prompt only | Low, product is invented | Up to 15 s | Mood and lifestyle B-roll with no visible pack |
| Image to video, start frame | 1 product image + prompt | High in early frames, can drift late | Up to 15 s | Reveals, rotations, pour and splash shots |
| Start and end frame | 2 images + prompt | Highest, both ends pinned | Up to 15 s | Before and after, open and closed pack, colour variants |
| Omni multi-shot storyboard | References + per-shot settings | High when references are clean | Up to 15 s in total | A short sequence with several angles in one generation |
Clip lengths come from the launch announcement. Exact durations, resolutions and credit costs depend on the tier and on where you run the model, so check the official Kling AI app or your provider's documentation before you plan a batch.
How to prompt Kling 3.0 for ecommerce product shots step by step
Here is the workflow for a single product beat. Run it once per shot on your shot list.
- Pick one shot from your brief and write its purpose in a single line, for example "show the matte finish of the lid".
- Prepare the start frame. Use a sharp, evenly lit photo or render of the real product, cropped to the aspect ratio of the placement you are making it for.
- Clean the start frame. Remove dust, reflections of the studio and stray props, because the model treats everything in the frame as something it may move.
- Decide whether you need an end frame. If the product must finish in a specific state (lid off, box open, a second colourway), prepare that image at the same size.
- Write the camera line first: one move, one speed. "Slow push in", "orbit 90 degrees left", "locked off".
- Write the light line: one change or one quality. "Soft window light from the left, a highlight slides across the lid."
- Write the action line: what moves, and how much. "A single drop of water runs down the side of the bottle."
- Add an anchor line that tells the model what must not change: "The bottle stays upright and unchanged, label fully readable."
- Add a negative prompt if your provider offers one, listing the failures you have seen: "warped label, extra caps, melting, text artefacts, hands".
- Set the duration to 5 seconds for a first pass and generate two or three variations.
- Review each clip frame by frame at full size, checking the label, edges and the last second, where drift usually appears.
- Keep the best take, note what failed in the others and change one line of the prompt at a time before the next round.
Writing the prompt as camera, light, action and anchor, in that order, is the core of the method. It reads the way a director briefs a camera operator, which is also how our guide to briefing an AI model like a creative director frames it.
What a good Kling 3.0 product prompt looks like
Here is a worked example for a skincare serum on a stone plinth, start frame only, 9:16, 5 seconds:
Slow push in from a medium shot to a close-up. Soft morning light from the left, a thin highlight travels down the glass. A single drop of serum falls from the dropper back into the bottle. The bottle stays upright and unchanged, label sharp and readable. Shallow depth of field, clean neutral background.
Negative prompt: warped label, duplicated bottle, hands, extra text, melting glass, flicker
Notice what is missing: no description of the bottle's shape, colour or brand. The start frame carries that, and describing it again gives the model a second, conflicting version to blend with yours.
For a start and end frame shot, the prompt describes only the transition:
The lid lifts straight up and moves out of frame to the right. Camera locked off. Even studio light, no change in exposure. The jar stays centred and unchanged.
More prompt patterns for specific categories are in our list of the best prompts for Kling AI product videos.
How to use the multi-shot storyboard for a product sequence
The multi-shot storyboard in Video 3.0 Omni lets you set duration, shot size, perspective and camera movement per shot inside one generation. For product work this is useful when you want three angles of the same product that match in light and colour, for example a wide shot on the plinth, a macro of the texture and an overhead of the open pack.
Keep each shot to one idea and give every shot the same anchor line about the product. If one shot breaks, regenerate that beat as a single image-to-video clip and replace it in the edit rather than rerunning the whole storyboard.
When to use native audio on product shots
Native audio means the model generates sound together with the picture rather than you adding it afterwards. On product shots it works best for diegetic sound, the sound the object would really make: a cap clicking, liquid pouring, a shoe hitting tarmac. Prompt it in the same plain way, for example "a soft click as the cap closes".
Voiceover and music are easier to control as a separate pass, because the line and the mix change per variant. Our guide on adding voiceover to AI generated video ads covers that step.
What we check before a product shot goes into an ad
This is Adsome's own working checklist, the review we run on every generated product clip before it reaches the edit. Adsome is a film and ad production studio in Stockholm that makes AI-produced ad creative for brands that spend on paid social, and these are the failure points we look for most often:
- The last second. Product drift tends to build up over the clip, so the final frames get the closest look, not the first.
- Label and logo at full size. Text that looks fine in a phone preview often turns out to be near-letters when paused at full resolution.
- Edges and contact points. Where the product meets a hand, a surface or liquid is where shapes melt or merge.
- Scale against props. A jar that grows or shrinks relative to the plinth reads as fake even when the label holds.
- Colour against the product page. We compare a paused frame with the hero image on the shop, because a returns problem starts when the ad and the parcel do not match.
Any clip that fails one of these points is regenerated or cut short before the problem appears.
Which fits your situation
- You have studio packshots already. Use image to video with a start frame. Your existing photos become the anchor for every clip.
- You sell in several colourways. Use start and end frames, or generate one clip and recolour from a new start frame per variant, so every variant keeps the same motion.
- You need a short sequence for a 15 second ad. Try the Omni multi-shot storyboard, and fall back to single clips for any shot that breaks.
- You only have a phone photo. Clean it up first, ideally with an image model, before you animate it. Our AI product photography tutorial shows how to get from a phone shot to a usable frame.
Common Mistakes
- Describing the product in the prompt. It competes with the start frame and causes blends. Describe the motion, not the object.
- Asking for several camera moves in one clip. "Orbit, then push in, then tilt up" usually produces a confused move. Split it into separate shots.
- Starting from a busy image. Props and clutter get animated too and pull stability away from the product.
- Going straight to 15 seconds. Long clips give drift more time to build. Prove the shot at 5 seconds first.
- Judging on a phone preview. Review at full size and paused, especially the label and the final frames.
Want the right model picked for every ad? Adsome makes performance video and image ads from your real product.
