During the 2026 edition of the I/O conference, the search giant revealed a new tool for the Flow ecosystem, giving users the ability to generate virtual replicas of themselves to act in audiovisual productions guided by artificial intelligence. Focused on content producers who seek to optimize recording time without having to set up lighting scenarios or face the lenses, the mechanism had its first public exhibition led by Elias Roman. The vice president of product management at Google Labs used his own synthetic identity to star in several algorithmically generated demonstrations during the event.
Step by step to generate a virtual identity in the application
To bring the cyber clone to life, the procedure requires just a few minutes and a smartphone. The user begins configuration by scanning a QR code that directs them to profile adjustments in Flow. From that point on, the software asks the person to recite specific number sequences out loud and move their face in various directions. This calibration step allows the technology to map vocal cords and facial biometrics from different angles, building a highly accurate and personalized three-dimensional model.
All this visual engineering runs on the OmniFlash engine, the company’s latest video architecture that retires the old Veo system. The big difference with this structural update is the stability of the elements during reproduction. While past generations suffered from visual hallucinations, where features melted or changed drastically between camera cuts, the new artificial intelligence is able to maintain the character’s anatomical cohesion from the beginning to the end of the recording.
Technical advances in image rendering with the OmniFlash engine
At the heart of this new functionality, OmniFlash represents a generational leap in the way machines process moving pixels. The company emphasizes that the focus of development was to eliminate the artificial and unstable appearance that marked synthetic clips until last year. With the drastic reduction of facial anomalies and texture mutations, the final result is much closer to a professional capture made in a traditional studio.
To democratize the use of the tool, the interface now accepts instructions in plain text, allowing anyone to change sets or costumes just by typing what they want to see. The platform also gained an intelligent routine automation system, where the producer can save lighting and style parameters in specific folders. This organization facilitates the application of the same visual identity in multiple future projects, optimizing the routine of those who publish material daily on social networks and expanding the use of AI for laypeople.
Traceability and security seals against digital disinformation
Knowing the risks inherent in voice and image cloning, the corporation embedded strict security locks in the source code to monitor the ecosystem. Any material exported by OmniFlash automatically carries the invisible SynthID signature, a cryptographic protocol created to reveal the synthetic origin of the file. Originally launched in 2023, this watermark imperceptible to human eyes already sweeps search engine results and has confirmed integration for the Chrome browser in the weeks following the event.
Another front of defense is the expansion of the C2PA standard, a global coalition that stamps authenticity metadata at the exact moment of media capture. First tested on the Pixel 10 smartphone lenses, the verification mechanism will be extended to videos recorded on the Pixel 8 and 9 lines very soon. The market is also waiting for Meta platforms, such as Instagram, to start reading and displaying these certificates of origin, creating a unified barrier against manipulation on the internet.
Restrictions on use and the continued fight against fake videos
Unlike what the market saw with the extinct Sora project, Mountain View’s current bet has an ethical limitation programmed directly into the database: you can only clone yourself. The interface summarily blocks any attempt to generate third-party replicas at the current stage of development. This architectural lock was the solution found by engineers to prevent the tool from becoming a factory of malicious deepfakes or unauthorized identity theft.
The crackdown on audiovisual counterfeits also encompasses other initiatives parallel to the invisible brand ecosystem:
- Explicit visual warnings in the Gemini app and other branded services whenever media is algorithmically constructed.
- Reverse engineering tools built into Google Lens and Search AI mode to check the veracity of a suspicious image.
- Adoption of the SynthID protocol by Nvidia cards and servers, tracking content directly on the hardware.
- Partnerships signed with giants such as OpenAI, Kakao and ElevenLabs to standardize reading and support for these security seals.
The constant improvement of these verification filters shows that the technology industry is trying to keep up with digital fraudsters, while unlocking new creative capabilities for Flow. By combining a high-capacity virtual studio with sophisticated provenance trackers, the company is trying to establish a commercial standard where algorithmic innovation can exist responsibly, without destroying the public’s trust in what they see on screen.
