Gemini Omni Flash is a multimodal generative AI model developed by Google DeepMind, specifically designed for high-resolution video creation and conversational editing. Released in May 2026, it serves as the foundational model for the Omni family, succeeding the Veo series. The model is built as an "any-input-to-video" generator, capable of processing combinations of text, images, audio, and existing video clips to produce cinematic content.
A primary feature of the model is its support for multi-turn conversational editing, allowing users to modify scenes, swap backgrounds, or change character wardrobe through natural language dialogue. To maintain visual fidelity, it can utilize up to five reference photos to ensure characters and locations remain consistent across frames. The model also natively generates synchronized audio and voiceovers, integrating sound directly into the video generation process rather than applying it as a post-processing step.
Technically, Gemini Omni Flash is a transformer-based architecture optimized for the "Flash" tier, prioritizing generation speed and cost efficiency for high-volume workflows. It is grounded in advanced physical reasoning to simulate realistic world dynamics, such as gravity and momentum. The model currently generates clips up to 10 seconds in length and includes integrated safety features like SynthID digital watermarking and C2PA Content Credentials for media provenance.