Gemini Omni 1.1 Flash

Gemini Omni 1.1 Flash is a multimodal video generation and editing model developed by Google DeepMind. Released as an update to the original Omni Flash, this version emphasizes precise creative control, enabling users to generate, extend, and refine high-resolution video through text, image, and video inputs. It operates as part of the broader Gemini ecosystem, utilizing a transformer-based architecture with native support for synchronous audio generation and advanced understanding of physical motion.

A primary feature of the 1.1 update is scene extension, which allows the model to analyze up to 10 seconds of prior video context to generate consistent continuations, reaching a cumulative length of 40 seconds. The model also introduced first-and-last-frame interpolation, a keyframing capability where developers specify the starting and ending images of a sequence to dictate transitions and camera movements. Output resolutions range from 360p draft previews—designed for cost-efficient iteration—to upscaled 4K final exports.

Technical Capabilities and Architecture

The model is built for an "any-to-any" workflow, meaning it can process disparate inputs (text strings, audio files, or static images) to produce a single cohesive video output. Unlike earlier video models that often struggled with physics, Gemini Omni 1.1 Flash leverages Gemini’s core world knowledge to simulate gravity, momentum, and fluid dynamics with greater fidelity. Its conversational editing interface allows for iterative refinements, such as swapping backgrounds or changing wardrobe, while maintaining character and lighting consistency across turns.

Prompting and Best Practices

For optimal results, prompts should include specific details regarding scene composition, lighting, and camera dynamics (e.g., "low-angle tracking shot" or "warm sunset lighting"). Because the model generates audio natively, users are encouraged to add sound descriptions at the end of their prompts to guide the integrated AAC audio track. Utilizing the 360p draft mode for initial compositions is recommended to verify narrative flow before committing to high-resolution 1080p or 4K upscaling.