Google's Gemini Omni video model leaks days before I/O 2026
A leaked multimodal video model suggests Google's next-gen AI is already live...
Deep Dive
Just days before Google I/O 2026 starts on May 19, a new video model called Omni has appeared inside the Gemini app. The picture comes from two separate leaks that happened nine days apart.
Key Points
- Two separate leaks nine days apart reveal Omni inside Gemini; supports up to 10 minutes of video at 30 fps
- Model includes temporal attention for object tracking and real-time transcription in 12 languages
- Generation capabilities confirmed (text-to-video up to 30 seconds), plus Workspace integrations for Meet and Drive
Why It Matters
Omni could make video as searchable as text, unlocking new workflows in compliance, education, and media production.