Grounding on your product and holding a face

Why a reference pack matters here

When your project has an approved reference pack, the generator passes your real product shot, your presenter, and your background into the render as named references, and the product reference is told to reproduce its actual shape, packaging, colour, and label text. A reference pack is like handing a stand-in actor your product off the shelf before the shoot, instead of describing it to them and hoping. It is the difference between a clip that shows your real product and one that shows a generic look-alike.

With no approved product image, the engine has to build from the product text alone. That is the usual reason a clip comes back looking generic: there was nothing solid to anchor it to. So if your product looks wrong, check the reference pack before you blame the brief. Reference packs are set up per project and have their own chapter.

Holding a person across the clip

Keeping the same face on screen is genuinely hard for text-to-video, and the tool guards it in two ways. It renders the clip as one continuous take rather than separate shots assembled later, and it re-renders each presenter photo into a multi-angle character sheet before the model ever sees it, which also helps the clip clear the provider's real-person check on input images.

Now the honest limit. These are guards, not a guarantee. Identity can still drift across a longer clip, and it drifts more across separate runs, so the same presenter in two different videos will not come out pixel-for-pixel identical. Treat consistency as something the tool fights hard to hold, not something it promises. Shorter clips and a clean, well-lit presenter photo in the pack give you the best odds.