Google is expanding the creative side of its Gemini platform with the introduction of Gemini Omni, a new multimodal AI model designed to generate and edit content from combinations of text, images, video, and audio inputs. The first release in the new Omni family, Gemini Omni Flash, is rolling out to the Gemini app, Google Flow, and YouTube Shorts.
The announcement at Google I/O 2026 builds on last year’s debut of Nano Banana, Google’s image-focused AI system that handled tasks like photo restoration, sketch-based design concepts, and visual ideation. Gemini Omni takes that same “what if we just let the AI cook?” energy and pushes it into video creation.
At launch, the focus is squarely on video generation and editing, with image and audio output support planned for later.
Video Editing Through Natural Conversation
One of the more interesting aspects of Gemini Omni is how it approaches editing. Rather than relying on timelines, keyframes, or menus buried three clicks deep like a treasure hunt designed by raccoons, Omni uses conversational prompts to make changes.
Users can modify videos through successive natural-language instructions, with the model maintaining scene continuity, character consistency, and environmental details between edits.
Google demonstrated examples where users transformed sculptures into bubble-like structures, altered interactions with mirrors to create liquid ripple effects, and progressively refined a violin performance scene across several prompts without losing the original context.
The company says Gemini Omni is designed to “remember” previous edits and build upon them, which could make iterative creative work less rigid and more exploratory.
Built Around Gemini’s Existing Knowledge Model
Google positions Gemini Omni as more than a visual effects engine. The company says the model combines Gemini’s broader reasoning capabilities with video generation to produce scenes that follow real-world logic more closely.
That includes handling concepts tied to gravity, motion, fluid behavior, and object interaction. One showcased example featured a marble navigating a complex chain-reaction track in a continuous shot, aimed at demonstrating smoother physics simulation and scene coherence.
The model also leans on Gemini’s wider knowledge base for more contextual creative prompts. In one example, Omni generated an alphabet-themed video where each letter corresponded to a specific object, complete with matching lower-thirds and stylistic formatting instructions. Another example created a claymation-style explainer about protein folding using stop-motion aesthetics.
That broader contextual understanding could make Gemini Omni particularly useful for educational content, social media clips, visual explainers, and concept prototyping. It feels less like a traditional text-to-video generator and more like a creative collaborator that occasionally speaks fluent storyboard.
Multiple Inputs, Single Workflow
Gemini Omni supports combinations of text, images, video, and audio references as inputs. At launch, audio support will begin with voice references, with broader audio capabilities expected later.
Google says the goal is to let creators start from nearly any source material and generate cohesive video outputs from it. That could include remixing existing footage, creating stylized reinterpretations of scenes, or building entirely new visual concepts from mixed media references.
The rollout across the Gemini app, Google Flow, and YouTube Shorts hints at where Google sees the strongest early demand: creators, short-form video production, and fast-turnaround content generation.
A Broader Push Into AI Creativity
Gemini Omni arrives as AI-generated video tools continue moving from experimental demos into consumer-facing products. Google’s approach leans heavily into multimodal flexibility and conversational workflows rather than standalone prompt generation.
For creators already using Gemini tools, Omni appears positioned as the next layer in Google’s growing AI ecosystem, connecting reasoning, editing, and content generation inside a single platform.
