Give each second one job before generating a product video.

Choose the production route before timing the story

First write the fact the viewer should understand: the package shape, a source-visible material or a verified construction detail. A generated action is not proof that a product works. Use recorded evidence when the message depends on a real preparation, result or performance.

A completed plan is not yet a film. Generating a preview does not make, replace or add a shot in the complete TVC. Keep those assets and task identifiers separate.

Allocate the message to the available shot intervals

The complete-TVC plan assigns the chosen total duration across the chosen shot count in near-equal intervals. The current storyboard editor lets you revise shot size, camera, description and proposed narration or sound; it does not expose a separate duration control for each shot. Do not promise a long final hold merely by writing a longer time into a description.

Fictional planning example, not an executed film: a 10-second, 9-shot pouch sequence needs to establish the actual pouch, direct attention to supported details and finish on a useful product view. The average allocation is about 1.11 seconds per shot, with rounded boundaries. That is a planning calculation, not observed model timing. Each cell needs a simple job; nine unrelated demonstrations would be difficult to read.

With 16 or 25 shots, the same final duration leaves less time per shot. Use the count that the message can support; more panels are not evidence of a better advertisement. If the story needs unequal editorial holds that the current route cannot express, change the plan or make a separate external edit from reviewed footage.

Prepare the product and any people before animation

Provide the actual product source and verified facts. When selected adult lead references are part of planning, keep their roles distinct from the product. They can guide the storyboard imagery; a four-view identity sheet is reference material, not four people to place in the scene.

Review the plan, generate the four grid candidates after checking that stage’s quote, and inspect every cell before selecting one grid. Check product edition, original printing, color, component count, source-compatible angles, people and contact. A good-looking grid can still contain an invented handle or a different person.

The complete route splits the selected grid into scene frames. Each shot video uses its corresponding reviewed scene frame, already containing the intended product and any person. The current H3 video input does not independently receive raw identity sheets and separate product photographs to reconstruct that composition. If a required face, hand position or scene is absent, resolve it in the image stage first. See the model-reference workflow for that boundary.

The selected 16- or 25-panel sheet provides fewer native pixels per cell than a 9-panel sheet of the same overall size. Enlarge the selected grid and inspect each panel at its available native detail. If necessary, download the grid and inspect or crop it in an external image editor. The current review screen does not provide a separate extracted-frame preview. Enlarging a crop cannot reconstruct unreadable lettering or supply missing product evidence. Use the missing-angle review when a frame would reveal an unsupported surface.

Check each stage’s quote, not an old per-second example

Planning, grid images, full shot production and a later retake are distinct stages. Review the current quote and actual selected model before approving each stage. A free local brief document does not grant generation credits or account access.

The edited 5-, 10- or 15-second film is not billed simply as that many generated seconds: the complete route produces the source clip for every planned shot. The grid stage also has its own image quantity. Use current pricing and account access and the actual stage quote instead of a fixed historical rate.

A quote is not a promise of acceptable content. Keep task status and visual acceptance separate, and check an uncertain existing request before submitting again. A retake is a new quoted operation; a technically completed but visually unsuitable result is not automatically a confirmed service failure.

Review the automatic edit and finish communication separately

For a complete TVC, the media step takes the planned interval from each source clip, orders the shots and creates a 720p, 30fps silent final edit. It removes source audio. This is automatic assembly, so the former advice to assemble every multi-shot ad outside CartVisual no longer describes this route.

Watch the whole film and inspect every cut, moving label, hand, reflection and product edge. Automatic trimming cannot correct a changed product or make an unsupported demonstration true. If one scene is wrong, identify the shot and check its available retake action and quote rather than assuming a new preview updates the film.

The optional single-shot preview uses one original approved scene image and its own selected duration/model settings. Raw provider output may contain audio or differ slightly in duration. Its output is not the complete film’s guaranteed silent edit; listen and inspect before reuse. Keep licensed sound, exact captions, voiceover, disclosures and any additional external edit as separate decisions.

Runway’s Image to Video Prompting Guide separates the image’s composition from the prompt’s motion and warns that source motion cues can conflict with a requested action. That is a useful vendor principle, not evidence that CartVisual uses Runway or guarantees the same control. Start from the source-visible action the frame can support.

Finally inspect the destination crop, current placement requirements and copy readability. Save the approved plan, grid, film task and published derivative under related identifiers. The placement preflight is a separate check, not an outcome inferred from generation.

Use the free brief builder for one motion direction

The free photo and video brief builder formats a local text draft. Its video form describes one shot, a camera action and a whole-second duration from 4 to 15. It does not inspect a photograph, call a generation model, create grid images or assemble a film.

Use it to express a source-supported movement, then adapt that direction to the actual route and interval you selected. Its duration field does not describe all current TVC choices, and its single-shot output is not a site-wide restriction. A copied eight-second direction is not an eight-second slot inside a ten-second, nine-shot film.

For a pouch label, a restrained push within the visible view is a narrower request than an orbit revealing the unseen back. Review the image first and keep exact promotional text outside generated pixels. The builder’s preservation language is a request to check, not a completed output test.