Google: gemini-omni-1.2-flash

Google: Gemini Omni 1.2 Flash is a multimodal video model from Google that takes combinations of text, images, audio, and video as input and produces video as output. It generates video from text and images, and it can edit existing video through natural-language conversation, so creators can refine a clip iteratively instead of restarting the entire creative process. Reference-driven creation lets you supply images of characters, scenes, or drawings to steer the result. The model shows an improved intuitive understanding of physics, including gravity, kinetic energy, and fluid dynamics, and it draws on Gemini's knowledge of history, science, and cultural context. That combination makes it suited to social ad drafts, product launch clips, e-commerce motion ads, app demo videos, seasonal campaign variants, performance creative testing, short influencer-style mockups, trailers and gameplay previews, and explainers built from short prompts. Gemini Omni Flash launched in public preview and is available in Google AI Studio and the Gemini API. At launch it generated 10-second clips, and the API initially did not support audio references or scene extension. Character consistency was limited in some scene changes and panning movements. On Video Arena it holds first place overall with an Elo of 1404, a 101 point gap ahead of Seedance 2.0 Mini.