At its I/O 2026 conference, Google introduced a new family of generative AI models called Omni. This technology, which developers describe as an “anything to anything” model, promises the ability to transform any input format – text, photo, or video – into entirely new audiovisual content. The first available version is currently the Omni Flash model, which is integrated into the company's Flow platform for video generation and editing.
The new Omni model brings several fundamental innovations compared to its predecessor, the Veo model. It allows users to upload their own video footage and use it as a basis for further generation in combination with text prompts. Google states that the model has a better understanding of the physical laws of the real world and can better maintain character consistency across different scenes, which used to be a weakness of older generative tools.
“Some of the clips were very good – much more consistent than when I tested the Veo model five months ago. Nevertheless, technical glitches still appear, where a character suddenly changes orientation in space or objects in their hands change their appearance during a shot,” says Allison Johnson, editor at The Verge, who tested the model.
In addition to creating fully animated scenes, Omni focuses on so-called deepfake technologies and real video editing. Users can add previously nonexistent elements to footage of a real person or change the background. In tests, the model was able to convincingly place a character into an airplane cockpit or in front of the Eiffel Tower, with the resulting level of realism being high enough to confuse even family members of the testers.
Despite technological advancements, the tool encounters practical limits in terms of cost and user-friendliness. The tool's operation is based on a credit system within the AI Pro subscription (20 dollars per month). One round of edits costs 40 credits, while generating a scene costs 15 to 40 credits depending on the length. With the current error rate, where the model often requires repeated instructions to achieve the desired result, creating a short video can thus become financially inefficient.
With the Omni model, Google confirms the trend towards absolute media synthesis, where the line between authentic recording and artificially created reality becomes imperceptible to the average eye. Although the technology is in the “uncanny valley” (a feeling of unnaturalness in almost perfect imitations), advancements in editing logic and working with text prompts suggest that Google is striving to create a comprehensive production tool for the general public.