The short answer
Use Kling 4.0’s dedicated Keyframes entry when approved images define both the opening and closing packshot; choose Omni Reference when separate assets need distinct roles. This documentation-based Kling AI image-to-video prompt guide is not a hands-on test, and the exact controls depend on the mode shown in your account.
Should Your Product Shot Use Kling 4.0 Keyframes or Omni Reference?
The distinction is what you can approve before generation. Kling’s 4.0 user guide describes Keyframes as a dedicated timeline entry and Omni as a reference-led entry; Omni also offers keyframe-to-video. Kling says the two primary entry points cannot be combined in one task. These examples treat them as separate routes, so check the mode label in your account before using either.
eyframes for a Defined Closing Packshot
Choose the dedicated Keyframes entry when approved opening and closing images show the same SKU—for example, an amber dropper bottle on a counter and then centered in the campaign layout. Your prompt describes the transition between known images instead of asking the model to invent a package view.
If the closing packshot exists only as an idea, make that still first or use a single-image route. A text instruction cannot provide the exact unseen label or cap. The first keyframe also determines the output ratio, so prepare it for the placement rather than assuming a square source will protect a vertical label.
Omni Reference for Assets With Separate Roles
Choose Omni when one image defines the product and another defines the setting, composition, or motion. Kling assigns uploaded images identifiers such as @Image1 and @Image2; explain each role so the model does not have to guess what to preserve. This route is a poor fit when references show conflicting package versions. Fix the source set first.
Prepare Product Assets for the Chosen Entry Point
Start with the exact SKU and an image cleared for AI processing and advertising. Record the rights owner, permitted channels, campaign period, and any supplier artwork or people shown. Kling’s current user terms make the uploader responsible for authorization and treat uploads as non-proprietary and non-confidential. Resolve the policy before uploading unreleased or confidential packshots.
Choose an image where the label, cap, silhouette, and color are easy to inspect. For Keyframes, use matching endpoint images and a plausible camera path. For Omni, assign @Image1 to the product and @Image2 to the setting. Each reference should have one job.
Set placement, ratio, duration, resolution, copy area, and sound before generating. A square image may not leave room for a vertical crop. TikTok’s current in-feed ad specifications list both formats and note that safe areas vary by format and caption length. A supported ratio cannot guarantee label placement.
For a fair comparison, hold SKU, claim, placement, duration, and review criteria constant. The two Kling AI image-to-video prompt examples below keep that goal while using different route-specific inputs. Log those inputs, so source-image effects are not mistaken for mode behavior.
Write a Keyframe Prompt for a Controlled Transition
Upload approved first and last images in the dedicated Keyframes entry and describe the path. Keep camera movement modest when label fidelity matters; a large rotation asks the model to invent unseen surfaces. Kling AI’s image-to-video prompt guidelines favor concrete language and warn that long negative instructions can increase errors. Describe the desired action; reserve exclusions for essentials such as unwanted subtitles or music.Prompt example:
0–2 seconds: Begin on the approved first frame. Keep the amber dropper bottle upright on the pale stone bathroom counter as the camera makes a slow, smooth push-in. Keep the front label facing the camera and readable. 2–4 seconds: Continue the push-in with the bottle centered; keep the warm side lighting, bottle shape, dropper, label, and color consistent. 4–5 seconds: Settle into the approved closing frame and hold the packshot steady. No subtitles or background music.
Use this only if the two images support that path. If the closing frame changes the angle or lighting sharply, resolve that visual contradiction before generating. I would check the label before the lighting; a polished transition is no help if it changes the SKU’s visible claim.
Write an Omni Prompt With Explicit Asset Roles
Use Omni when references contribute different things. State what each image controls, the camera move, and the product details that stay visible. Kling favors direct descriptions over abstract style labels.Prompt example:
@Image1 defines the approved amber dropper bottle: preserve its silhouette, dropper, front label, color, and proportions. @Image2 defines only the warm bathroom counter, pale stone surface, and side lighting. Create one continuous five-second 9:16 shot. Keep the bottle upright and front-facing as the camera slowly pushes in. Keep the full label inside the frame and leave clear space above it for approved campaign copy. No subtitles or background music.
Use references that support the shot. For tiny legal copy, start with the highest-resolution approved image and inspect the frames; a prompt cannot replace legible source pixels.
Check Packaging and Scene Continuity in Each Result
Review at normal speed, then inspect the opening, middle, and final frames at delivery size. Check label and claims, package shape, orientation, props, background, and crop. Mark defects with timecodes and decide whether to trim or revise the source. I want to see what survives a second look.
Keep one record row per attempt: model and entry mode; approved source files and rights; prompt version; ratio and duration; credits shown and result status; defects and timecodes; editor minutes; final decision. Kling’s Credits Policy says credits are refunded when a generation fails. A completed clip that fails brand review is a different outcome.
This is why credit price is not the same as usable-ad cost. Count the generations behind an approved version, then add review, editing, copy compositing, captions, resizing, and reruns. The next edit is part of the cost. Compare the routes by the clips that pass the same product-fidelity review, not by the cost of one render.
Which Kling Setup Should Your Team Revise or Switch?
When motion works but the package drifts, reduce camera travel or improve the source. If endpoints conflict, revise the closing image. If references blur together, simplify Omni and clarify roles. Record the exact mode and model because controls vary by rollout.
Wizstar’s public AI Video Agent page lists Kling 3.0 Omni, not Kling 4.0’s dedicated Keyframes entry. These 4.0 examples apply to Kling’s native interface when available. If the product must remain the visual anchor as a scene is built around it, consider a product-led workflow and verify which model the platform currently offers. I have the documentation, not a finished test.
Conclusion
Use Keyframes when approved endpoint images define the shot; use Omni when separate references need distinct roles. Hold the product facts and review criteria steady, then judge the exported crop and package—not just the preview. Start with the constraint your team can approve today, and keep the attempt record with the ad.
FAQ
- Can a team share a Kling draft without sharing an account login?
- Yes, if the account offers Team Space: invite collaborators under their own accounts and assign roles. Kling’s terms separate team and personal spaces and say Team Assets belong to the Team Owner unless law or a written agreement says otherwise. Confirm ownership before client work.
- Can a square product image become a vertical clip without cropping its label?
- There is no setting that guarantees it. Start with a vertical-safe source or leave room around the label, then inspect the exported frame against the placement’s safe area.
- Are failed Kling image-to-video generations charged credits?
- Kling’s current policy says credits are refunded when a generation fails. It does not say a completed but unusable clip counts as a failure, so log technical errors separately from creative rejections.
- Can a brand keep its uploaded product images out of training data?
- The public terms do not promise a blanket exclusion. They allow input processing for model improvement or training, provide a support contact to request revocation, and say not to upload confidential content. Keep unreleased images out unless a separate written agreement covers them.
- Do Kling’s current terms cover paid ads made from supplier photos?
- Supplier permission for AI processing and ad use is still required. Output rights depend on the account: Kling’s general terms require written permission for commercial output and stipulate Kling branding in some cases, while its paid-service terms grant members commercial-use benefits subject to exceptions and the subscribed plan. Get written confirmation for the actual account and campaign before launch.
