naiwa 2026-05-20

Videos can now be generated from any data.

At the Google I/O 2026 conference, DeepMind CEO Demis Hassabis announced the new multimodal Gemini Omni lineup, focused on combining different types of content generation.
The first model is Gemini Omni Flash — a system capable of creating videos with sound accompaniment based on various input data, including images, audio, diagrams, and other videos.

According to Google, the model better understands the physics and logic of the real world, relying on Gemini 3.5 technologies.
An editing mode has also been implemented: users can modify generated videos using text commands while preserving key details. Real video processing is supported as well — for example, adding effects or changing styles without altering faces.

Additionally, a feature for creating digital avatars from photos and voice with subsequent video generation from text is being tested.

Access to Omni Flash is open to Google AI subscribers (Plus, Pro, and Ultra), as well as Flow users. For YouTube Shorts creators and the YouTube Create app, the model is available for free. Integration into the API is expected soon.
At the same time, the company introduced Gemini 3.5 Flash — a model focused on programming tasks and agent systems.
Video on YouTube 
youtube.comhttps://www.youtube.com/live/wYSncx9zLIU