Google unveils Gemini Omni Flash for multi-modal video generation

Omni Flash functions as a video-centric counterpart to Google’s Nano Banana image generator, which has produced over 50 billion images since its debut last year. While the current iteration limits output to 10-second clips featuring synchronized audio and video, Google DeepMind senior research director Dumitru Erhan confirmed that teams are already working to extend this duration. Product lead Nicole Brichtova noted that the model supports personal likeness integration, a feature that has seen high engagement in previous image-based models.

Unlike the text-to-video capabilities of Google’s existing Veo model, Omni Flash leverages the broader training data inherent to the Gemini ecosystem. Koray Kavukcuoglu, Google DeepMind CTO and chief AI architect, emphasized that this integration grants the model significantly more world knowledge than specialized video tools. Starting Tuesday, the technology will be accessible to users through the Gemini app, Google Flow, and YouTube Shorts.