Create a product-motion concept from an approved still
Anchor composition with a source image and review every frame for product drift.
Use a prompt with four parts: shot purpose, camera movement, environmental movement, and end state. For example: “Five-second product reveal. Slow clockwise camera arc of about fifteen degrees while a narrow highlight travels once across the bottle shoulder; background and surface remain still; settle on the original front-three-quarter view for the final second. Preserve bottle silhouette, cap proportions, label placement, label text, color, and tabletop contact. No new objects, no cuts, no morphing.” This is a review specification, not a guarantee that negative instructions will prevent drift.
The provider documentation for Kling 2.6 Pro image-to-video requires a prompt and a starting image URL. It publishes five- and ten-second options, with five seconds as the default, and also permits an optional ending image. Those provider controls explain the model boundary, but they should not be copied into an OfflineCreator MCP call unless the MCP schema itself exposes them.
Start with an approved still and one motion idea
A product-motion concept should test movement without reopening the approved composition. Begin with a rights-cleared source still whose crop, product angle, lighting, visible label, colors, and background have already passed human review. Duplicate that file for the experiment, record its filename or asset ID, and state one shot purpose such as “a restrained five-second reveal for a product-page teaser.” The source remains the visual reference; the generated clip is a concept that must be checked against it.
OfflineCreator Studio currently lists Kling 2.6 Pro Motion as its image-to-video model at 42 credits. Its template library describes a cinematic product-motion recipe as a five-second reveal and a bottle-orbit recipe as an upload-based image-to-video workflow. Those are useful boundaries for this task, not evidence that an arbitrary package, logo, reflection, or small printed detail will stay unchanged.
Reserve, upload, and submit in the published MCP order
Call `list_models` immediately before generation and choose the current image-to-video model ID rather than copying an old identifier. Then call `generate` with that model ID, a motion prompt, a supported aspect ratio, and a deliberate wait choice. For an image-to-video model, the public MCP implementation returns the reserved generation when an upload is required instead of polling as though processing had already started.
Next call `upload_input` with the reserved generation ID and provide the source image as either a local path inside the configured upload root or as base64 image data. Although both schema fields are optional, the client requires at least one: it uses `filePath` when that field is present, otherwise it uses `imageBase64`, and it rejects a request with neither. Supply only the intended field so the input choice is unambiguous. The tool accepts PNG, JPEG, WebP, or GIF content types, uploads the selected image, and then submits the reserved generation using `16:9`, `9:16`, or `1:1`.
Do not add a duration argument to either MCP call. The published `generate` and `upload_input` schemas expose no duration field, even though the underlying provider endpoint has duration choices. For the current Studio product-motion template, plan the documented five-second concept and confirm the live model record before spending.
Apply an editorial frame-by-frame review method
The first four categories below are an editorial comparison method for this concept, chosen to make human review repeatable. They are not model-performance facts, guarantees, measured defect classes, or categories supplied or validated by C2PA. The reviewer sets the acceptance threshold for the specific approved still and records observations without inferring how frequently any defect occurs.
- Geometry
- Compare silhouette, cap, corners, openings, and contact pointsEditorial check: compare sampled frames with the approved still and record bending, changing dimensions, or loss of surface contact as reasons for this reviewer to reject the concept.
- Identity
- Compare logos, labels, typography, colors, and distinctive marksEditorial check: pause on readable frames and reject this concept when a word, mark, color, or placement differs from the approved reference.
- Materials
- Track reflections, shadows, transparency, texture, and liquid levelEditorial check: record whether matte or glossy appearance, printed design, reflections, shadows, transparency, or liquid level changes relative to the source.
- Scene continuity
- Check the background, crop, camera path, and start-to-end transitionEditorial check: compare start, sampled middle frames, and end; note appearing props, changing horizons, unstable edges, abrupt acceleration, and an unusable final hold.
- Provenance
- Retain the source, prompt, model, generation ID, output, and review decisionIf the delivery system supports Content Credentials, inspect and preserve them, while keeping the local manifest because provenance metadata can be incomplete or removed.
Choose the next action from the review result
If the clip preserves the product but the movement is unclear, keep the same source still and revise only the camera or environmental-motion phrase. If identity, packaging, or geometry drifts, do not hide the defect with a fast cut; move to the drift-review workflow and decide whether a more restrained prompt or a different source angle is warranted. Use the approved-image animation guide when the task is a general still rather than a product-specific authenticity review, and return to the workflow directory if the deliverable changes.
Keep the page inside its evidence boundary
Treat the bottle prompt, comparison categories, and rejection rules on this page as an editorial working method for one short product-motion concept. They do not establish measured prompt success, exact product fidelity, reliable text rendering, production readiness, advertising performance, creator preference, defect frequency, or a customer outcome.
Before a paid run, the operator should recheck the live Studio model list, template, MCP schemas, and provider endpoint, then record the actual model, input, duration, output, and review decision. Keep research-coverage gaps and untested-method caveats in the evidence brief instead of presenting them as facts about model behavior.