How to combine real demonstrations with generated product footage

Give each source an evidence role

A real recording can document what happened in its recorded setup. It does not automatically establish every advertised use or a measured comparison. A generated scene can provide atmosphere or illustrate a verified appearance; it cannot become a new observation merely by sitting next to a real clip.

Imagine a fictional coffee-grinder edit. D1 is a real recording of a known unit grinding beans under a recorded setup. G1 is a generated kitchen shot with a perfectly even pile of grounds. If G1 follows the grinding action, viewers may read the pile as D1’s actual output. The file being labeled generated in an editor does not make that causal message safe.

Read the cut as a buyer would

Watch D1 → G1 without relying on your source notes. Ask what the sequence claims happened. “These are the grounds it produced” needs an actual supported result, not an attractive replacement. The same issue arises with a real wipe followed by an invented clean surface, or a real switch press followed by a generated performance effect.

Two useful corrections are different. Replace G1 with the documented result from D1 when that result serves the intended claim. Or change the sequence and message so G1 clearly serves a separate atmospheric role, without presenting a produced pile as measured output. If the misleading connection remains, hold the edit; a small AI label is not a substitute for changing it.

Keep source facts beside the edit decision

Keep product edition, source type and the source/result relationship in an external edit note. These identifiers and outcomes are hypothetical; no grinder or generation test was carried out for this guide.

Check the sequence, then its sound and copy

Adobe documents separate Source and Program Monitors for individual footage and the assembled sequence. Inspect both. Compare the real unit’s controls, label and color with every generated appearance. Correct identity still does not turn synthetic action or audio into a recording of that unit.

Review the export with its actual audio track. The current complete TVC assembly is silent; an independent raw clip can have different audio behavior, so inspect that file. Added motor sound, “quieter” narration or a timer can introduce a performance claim absent from the picture. Use sound-and-fact review before accepting the final mix.

Approve the message rather than the asset category

Creatify identifies invented features and inconsistent visuals as risks in its own agent description. That vendor account is background for a human review checkpoint, not independent proof of detection or a guarantee about CartVisual.

For the fictional E1, record which observed result the cut actually uses, which generated scenes remain and what each communicates. If copy changes from “a kitchen moment” to “uniform grounds every time,” reopen the claim review. Visible appearance and demonstrated performance need different evidence. Keep real source recordings and the reviewed final export; publishing a new edit is a new message decision.