Gemini Omni 1.1

Google's second video model here, and it is not a Veo. It runs on the Interactions API, which is what lets it do the one thing nothing else in this catalog does: take a clip you already have, read a sentence describing the change, and hand back that same clip with the change made. Everything else here answers by making a new take from scratch.

Every take arrives with its own synchronized sound — dialogue, effects, ambience — and there is no toggle for it, so sound is not a separate line in the budget. It works from a prompt alone, from a first frame, from a first-and-last pair it interpolates between, or from up to three reference images that guide who and what is in the shot without being frames of it.

There is no duration control at all: the model decides how long the take runs. The hold is sized at the eight-second ceiling and settles down to what the render actually cost, so a short take is not charged as a long one. That ceiling is 67 credits at 720p, 100 at 1080p and 200 at 4K — pick the resolution before you press, because this is one of the few models here where the top tier costs three times the base.