The zoom out from a person to the planet is one of the few genuinely durable AI video formats, and the reason is not that it looks impressive. It is that it gives a stranger a reason to keep watching that has nothing to do with the person in it.
A clip that begins with a face invites the viewer to evaluate that face, which is a strange thing to ask of a passer-by. The same clip that begins with a face and pulls back to a landscape is asking the viewer to be a witness. The subject changes from something they have an opinion about to something they are seeing, and the opinion goes away.
Why the effect works on a stranger with a photo
That is also the technical reason it works with generated video. The zoom is a continuous camera move with a stable subject, and the first and last frames are the two things the model has to get right. Everything in between is smooth motion that the model can interpolate, which is the part it is best at.
Where it fails is when the starting image is not a clear human figure against a clean background. A group photo, a heavily filtered selfie, or a subject against a busy wall all give the model the wrong kind of thing to hold onto while the camera travels, and the result drifts.
If you want to see how the effect behaves on different kinds of source images, this Earth zoom out AI video effect page walks through the input that works. The honest limitation is that it is a format for people and not for products, which is worth knowing before you build a product catalogue around it.
