Google released Gemini Omni 1.2 Flash on Nov. 5, an update to its AI model that makes and edits short videos with sound. People describe a scene in words, and they can add pictures of a character, a product or a style, a short clip of the movement they want, and sound samples. The new version makes clips of up to 15 seconds in one go, up from 10 seconds before.
Another big change is that people can fix a clip instead of starting over. They ask for one change at a time, such as a different product, background, lighting or weather. In one of Google's samples, the background behind two characters in armor changes from a striped tent to a forest, a stable, a market fair and a castle, while their performances stay the same. Google says faces, clothes, packaging and brand colors now stay the same as the camera moves, so a shop can turn its product photos into try-on videos and ads.
Google also says the characters talk more naturally, with lips that match the words more closely and a separate voice for each speaker. For the scene in the video, Google gave the model a picture of a glass conservatory, pictures of four characters and a short voice sample for each, and the model made a 15-second animated scene in which each character speaks in the voice it was given.
Developers and businesses can use the model now through Google's tools and its cloud service, at 30% less than the previous version. Anyone with a Google account can make 15-second clips at no extra cost in Google Vids, Google's app for putting together videos. The longer clips are also rolling out in the Gemini app to subscribers of Google's AI Ultra plan. Every video carries a hidden mark that people cannot see but software can read, so the clip can be identified as made by AI.
Most of what Google says about quality comes from its own samples. On an outside ranking built from people's votes, the new version ranks first both for videos made from a written description and for videos made from a picture. Those results are early, and it leads only narrowly (in the written-description list, just ahead of Google's own previous version), so its place could still change. Google says it is still working on fast sports action and complex physics, and on its cloud service, guiding a video with sound samples needs Google's approval for now.