Google has unveiled its latest generative AI model, Omni, which the company describes as an "anything-to-anything" system capable of transforming any type of input—whether text, image, video, or audio—into any other type of output. Currently, the model's primary application is video generation, and it is available through Google's Flow platform, previously used for its Veo model. Omni represents a significant step forward in AI's ability to create realistic, context-aware videos, but as early testing reveals, it is far from perfect.
The model's core innovation is its multi-modal versatility. Users can upload a video clip along with a text prompt to serve as the starting point for a new creation. Google claims Omni incorporates more real-world knowledge than its predecessors, leading to better character consistency throughout a video. To put these claims to the test, a senior reviewer conducted a series of experiments, starting with a familiar subject: a stuffed deer named Buddy, which had been used in earlier tests with the Veo model. The goal was to see if Omni could generate coherent, engaging videos of Buddy on vacation, complete with consistent character details and logical scene transitions.
The results were a mixed bag. Some clips were remarkably coherent, maintaining Buddy's appearance and the narrative flow of the prompt. For instance, a video of Buddy packing for a cruise ship vacation showed him placing a jar of honey in his suitcase, which later reappeared as a squirt bottle of honey he uses as sunscreen. This clever narrative beat demonstrated the model's ability to follow a storyline, but the visual consistency fell apart upon close inspection. The honey bottle changed shape and color multiple times, and the final frame appeared to be a garbled collage of previous elements. Such glitches are common in AI-generated video, but Omni's improvements over Veo are still tangible. The model handles text-based edits more effectively than its predecessor, though the results are far from seamless.
The reviewer also tested Omni's capacity for deepfakes by uploading a selfie video with a neutral expression and prompting the model to generate clips of herself eating spaghetti, sitting in an airplane seat, and posing in front of the Eiffel Tower with a baguette. The results were unsettlingly convincing. Her husband, who sees her daily, was fooled by the pasta-eating clip, noticing only that the bowl looked unfamiliar. The deepfakes displayed minor AI tells—a too-perfect clink of a fork, a background character appearing twice—but overall, they would likely deceive most viewers, especially on social media. This raises serious concerns about the potential for misuse, such as creating convincing fake videos of people without consent.
Omni is not free. Google operates a credit system where generating videos costs between 15 and 40 credits depending on length and complexity, with single edits costing 40 credits. The reviewer's $20-per-month AI Pro plan includes 1,000 credits, which were quickly depleted after generating about 20 clips and making a few edits. This pricing structure means that achieving a polished, vision-matching video could require significant financial investment, especially if users need multiple iterations. The model's tendency to introduce unwanted changes—like adding antlers to the stuffed deer that had none—further complicates the creative process. When prompted to remove the antlers from one scene, Omni complied but then added them to all other scenes, illustrating the unpredictable nature of these generative systems.
Generative AI video models have advanced rapidly. Just a year ago, creating a realistic video required extensive training data and computational resources. Now, tools like Omni democratize this capability, putting it in the hands of anyone with a subscription. This democratization brings both creative opportunities and ethical dangers. On one hand, artists and content creators can produce high-quality promotional videos, storyboards, or educational content with minimal effort. On the other hand, bad actors can manufacture fake evidence, impersonate individuals, or spread disinformation. The technology's maturity means that even experts may struggle to distinguish real from generated content without careful scrutiny.
Google's Omni arrives amid a broader wave of generative AI developments. At Google I/O 2026, the company positioned Omni as a cornerstone of its future AI ecosystem, emphasizing its ability to understand and generate across modalities. This aligns with industry trends: other tech giants are similarly investing in multi-modal models that blur the lines between text, image, audio, and video. For instance, Meta's Forum platform and OpenAI's Sora model (still in limited release) are competing in this space. The race is not just about quality but also about integration—making these tools seamless parts of everyday software like Google Workspace or YouTube. Omni's current integration with Flow is a step in that direction, but the experience remains clunky for non-experts.
The implications for journalism, entertainment, and personal communication are profound. News editors might use Omni to quickly visualize complex stories, while filmmakers could generate rough cuts of scenes before production. However, the ease of creating convincing deepfakes threatens to undermine trust in visual media. Policymakers are already scrambling to regulate AI-generated content, with some jurisdictions requiring explicit labeling. Google itself has implemented safety measures, such as digital watermarks and usage policies, but these are not foolproof. The reviewer noted that her deepfake videos, while imperfect, were good enough to fool her own family, highlighting the urgency of developing robust detection and authentication systems.
The uncanny valley remains a persistent challenge. Even Omni's best outputs exhibit subtle irregularities—a misplaced shadow, an unnatural gait, a flickering background—that betray their artificial origin. Yet these indicators are becoming less noticeable as models improve. The reviewer observed that while she was shocked by Veo 3's realism a few months earlier, Omni's capabilities felt both impressive and exhausting. The rapid pace of advancement means that the gap between generative AI and genuine human-created content is closing faster than society can adapt. Ethical guidelines, educational campaigns, and technical safeguards must evolve in tandem to prevent the technology from outpacing our ability to manage its risks.
For now, Omni is a powerful but flawed tool. It excels at short, simple clips with limited action but struggles with extended narratives and consistent object permanence. Its deepfake capabilities are frighteningly good, yet the model often introduces nonsensical elements that spoil the illusion. As Google continues to refine Omni—and as competitors push their own models—the line between real and generated video will only blur further. Understanding these systems' strengths and weaknesses is the first step toward using them responsibly. The key takeaway from this hands-on test is that generative AI is no longer a distant promise; it is here, wild and untamed, and it demands our attention as both a creative partner and a potential threat.
Source: The Verge News