Google presented a new avatar system for its Flow platform at the I/O 2026 conference, allowing users to create digital copies of themselves and embed them in artificial intelligence-generated videos. The feature was developed specifically for content creators who want to appear in productions without needing to be on camera. The demonstration was performed by Elias Roman, Vice President of Product Management at Google Labs, who inserted his own digital copy into different AI videos during the presentation.
Como the new avatar system works
The process of creating a digital copy is simple and straightforward. Users need to scan a QR code with their cell phone to access Flow account settings. Then, Google asks them to read numerical sequences out loud and turn their heads at different angles. The system analyzes the face and voice from multiple perspectives to generate the personalized digital copy.
The feature works in conjunction with the Google OmniFlash video model, which replaces the previous Veo model. Esse new model offers significant improvements in video production quality, especially in character consistency throughout scenes. Enquanto Previous versions featured characters that distorted or changed appearance between takes, OmniFlash solves this problem to a large extent.
The OmniFlash model and quality improvements
OmniFlash is the core of the new avatar system and brings substantial improvements over previous models. Google highlights that the main improvements focus on the overall quality of video production and maintaining character consistency between scenes. The technology reduces facial distortions and appearance variations that were common in previous generations of generative AI for video.
Adicionalmente, users will be able to edit scenes using natural language commands, making the process even more accessible. Google is also adding automated workflows and reusable command systems to Flow. Isso allows creators to automatically categorize similarly styled videos into folders and reuse the same content settings over and over again, expanding the reach of generative AI to a broader user base.
Recursos security and generated content detection
Google has implemented security measures to monitor platform usage. Todos videos created with the OmniFlash model feature the SynthID invisible digital watermark, designed to detect artificial intelligence-generated content. The SynthID system, introduced in 2023, is already actively working on the Busca and Google and will arrive on the Chrome browser in the coming weeks.
The company is also expanding its C2PA technology, which allows photos to be tagged the moment they are taken. Inicialmente used in the camera application of Pixel 10, the system will soon be active in video content on models Pixel 8, 9 and 10. Espera is also expected to support C2PA soon, creating a broader standard for content authentication.
Limitações current and future system
Embora At first glance the system may resemble the now discontinued Sora, the Google system initially only allows users to create their own digital copy. The option to create AI avatars of others is currently unavailable, a restriction implemented to prevent malicious use such as unauthorized deepfakes.
Measures against deepfake content are being expanded beyond SynthID:
- The Gemini app and Google services show users when content has been generated by AI
- Google Lens and Modo AI from Busca now offer the ability to identify synthetic content
- Nvidia has added SynthID support to its ecosystem
- OpenAI, Kakao and ElevenLabs started supporting the detection system
Google continues to improve its verification systems to facilitate the detection of deepfake content while expanding the creative capabilities of the Flow platform. Combining powerful creative capabilities with sophisticated sensing technologies strikes a balance between innovation and responsibility in the use of generative artificial intelligence.
