Google’s Omni is Here: The High-Stakes Gamble on ‘Anything-to-Anything’ Video Generation
Table of Contents
Beyond the Text Box: The Omni Ambition
Google’s latest foray into generative AI isn’t just another incremental update to a chatbot; it is an attempt to dismantle the barriers between different types of media entirely. Unveiled as part of the Google I/O 2026 cycle, the Omni family of models is designed for ‘anything-to-anything’ processing. In theory, this means a user could feed the model a photo, a snippet of audio, or a block of text, and ask for it to be transformed into any other medium. While the full scope of that vision is still unfolding, the first tangible manifestation has arrived in the form of Omni Flash, now integrated into Google’s Flow video editing platform.
For the average user, Omni represents a significant shift from its predecessor, Veo. Where Veo often felt like a tool for creating isolated, surreal clips, Omni attempts to anchor its generations in real-world knowledge and, more importantly, visual consistency. The goal is to solve the ‘hallucination’ problem in video, where characters or objects morph unpredictably from one frame to the next.
The Consistency Gap
Testing Omni in a real-world scenario reveals a tool that is simultaneously impressive and deeply flawed. In a series of experiments involving a consistent character—a plush toy—the model showed marked improvement over Veo in following complex narrative prompts. Google claims Omni understands physical world dynamics better, but the reality is a mixed bag of high-fidelity renders and jarring ‘AI jump scares.’
In one instance, a prompt for a playful montage of the character packing for a cruise resulted in a coherent sequence, including a narrative payoff where a jar of honey was mistaken for sunscreen. However, the visual stability faltered; the honey container shifted shapes multiple times throughout the clip, fluctuating between a jar and a squeeze bottle. The final frame of the sequence devolved into a chaotic smear of visual elements, a reminder that while the AI can ‘plot’ a story, it still struggles to maintain the physical identity of an object over a timeline.
The editing capabilities have also seen an upgrade. Users can now suggest text-based edits to existing AI clips. While Omni is more responsive to these commands than Veo 3, the results can be unpredictable. Attempting to refine a character’s facial expressions often resulted in an ‘uncanny valley’ effect, and specific requests to remove unwanted artifacts—such as misplaced antlers on a toy—occasionally caused the model to inadvertently apply those same artifacts to every other scene in the video.
The Deepfake Dilemma
Where Omni truly becomes unsettling is in its ability to integrate AI-generated content into real-world footage. By using a neutral selfie video as a base, it is possible to generate highly convincing deepfakes of oneself performing complex actions—such as eating pasta or standing in front of the Eiffel Tower.
The level of realism is high enough to fool those closest to the subject. In a blind test, a spouse was unable to distinguish between the real person and the AI-generated version of them eating pasta, noting only that the bowl looked unfamiliar. While technical tells remain—such as a slightly synthetic sound of a fork hitting a plate or repetitive background characters in an airplane shot—the ‘good enough’ threshold for social media deception has effectively been crossed.
The Cost of Creation
Despite the technical leaps, the barrier to entry for Omni is financial. The model operates on a credit system, with clips costing between 15 and 40 credits depending on complexity and length. For subscribers of the $20-per-month AI Pro plan, which provides 1,000 monthly credits, the burn rate is aggressive. A few dozen generations and a handful of edits can deplete a significant portion of the monthly quota.
This pricing structure suggests that Google views high-end video generation as a resource-intensive luxury rather than a casual utility. For creators seeking a precise vision, the costly back-and-forth required to fix AI glitches may make the tool less accessible than the marketing suggests.
