It's a beautiful afternoon at Dolores Park in San Francisco and I'm singing “Happy Birthday” to a prehistoric dinosaur. A pink cupcake magically appears as I sing, but it’s actually an AI-generated avatar of me. The content is cringeworthy, with some jumbled moments and nonsensical outfits, but that man in the video looks uncannily like me.

Google's Gemini app uses its new Omni video model to generate these digital clones. I paid $20 a month for access and quickly hit the usage limits. I tried two prompts: one where I sang to a dinosaur on top of a hill, and another where I surfed under the Golden Gate Bridge. The results were eerily similar to real footage.

The photorealistic background is impressive, with the AI seamlessly blending into the environment. However, the chaotic moments are jarring – a cupcake appearing out of nowhere or my voice stuttering on certain words. These deepfakes could be used maliciously, but Google claims it prioritises safety.

Despite the imperfections, these hyper-realised versions felt more real than listening to voicemails or watching fun weekend clips. It’s unsettling to see a digital clone of oneself always ready to do anything. What does this mean for our perception of identity in an increasingly tech-driven world?