Google showcases Gemini Omni through five creator demos, pushing video editing toward “conversational” AI

Google on August 7 highlighted how five builders are using its Gemini Omni video AI model in demos ranging from cinematic camera changes to voice-driven weather and sketch-to-video animation. The examples, published in a Google blog post, are designed to show that the system can not only generate new footage but also perform edits while keeping a coherent scene.

At the center of Google’s pitch is Omni’s ability to treat video creation and editing as an interactive conversation. The company says users can alter camera angles, swap objects, and convert simple sketches into realistic animations, with outputs that aim to look natural and smooth. Google also frames Omni as understanding how the world works, which it says helps keep edits consistent with real-world logic.

The post points to Gemini Omni Flash as a key step in the Omni family. Flash is described as Google’s first model in that family, introduced during Google I/O earlier this year. Google says Omni models can generate videos from text, image, video, or audio references, and that they can also edit a user’s existing videos, rather than requiring creators to start from scratch.

Google’s developer access update is another pillar of the rollout. According to the post, Google recently gave developers access to Omni, and builders have already used it for a range of projects. The company then illustrates the concept with creator-specific examples that emphasize scene continuity during edits.

One demo focuses on camera and perspective continuity. Builder Leon Lin, identified in the post as @LexnLin, captures a woman standing in a city from roughly 20 different perspectives, including close-up and long-distance views, straight-on and profile shots, and angles from above and below. Google says some shots zoom in while others hold steady, and that backgrounds shift accordingly, with the person remaining properly situated amid pedestrians, cars, trams, and buildings.

Other demos emphasize user control over content and timing. Builder Carlos Santana (@DotCSV) is described as changing an outdoor scene using voice input, shifting lighting from day to night, adding cloudy skies and rainy sounds, turning leaves orange, and covering the ground with snow. In a separate category, Google describes Omni in Google Flow, where sketch inputs can be turned into realistic video and users can decide whether the hand-drawn elements become part of the final clip.

In the sketch-and-character space, Pan (@sebatheepan) demonstrates transformations of everyday objects into playful animated scenes. The post describes a lemon becoming a submarine bobbing in the ocean, a cup of espresso turning into a hot air balloon, hot peppers becoming a sleeping dragon that breathes fire, a match recast as a rocket ship, and scissors turning into a shark searching for a snack. The underlying idea, as Google presents it, is that Omni can follow doodle guidance for how elements move.

Google also highlights style transfer and aesthetic changes without breaking the action. Builder Jerrod Lew (@jerrod_lew) is said to use Omni in Google Flow to render a woman walking down a street in four animation styles, with the clip transitioning fluidly from live-action to anime to claymation. Google claims the woman’s forward progression is maintained even as the visual style changes, which is positioned as evidence of scene coherence.

Beyond individual creative projects, the post includes business-adjacent use cases through Hyperagent (@hyperagentapp). Google says Hyperagent used Omni to visualize three concepts with video: layering landscaping into an empty park to create a before-and-after design proposal, animating a professor to explain business dashboards by personifying data, and gamifying a to-do list with a character clearing tasks. Google’s broader message is that Omni can blend real-world logic with creative editing, potentially turning static ideas into dynamic storyboards.

What Google does not provide in the blog post is detailed information on model performance benchmarks, pricing, contract terms, or the exact limitations of current edits. The company also does not specify which of these capabilities are available on every platform at the same fidelity level, though it lists multiple entry points, including the Gemini app, Google Flow, Google AI Studio, the Gemini API, and a Gemini Enterprise Agent Platform. Creators and developers may need to test for how consistently the system handles complex scenes, long-form continuity, and harder-to-predict edits.