Reece Rogers 在 2026 年 5 月 21 日试用 Google Gemini app 的 avatars 功能，付费每月 20 美元的 Google AI Pro 后，很快就把使用额度用完；该额度每 5 小时重置一次。他用约 5 分钟建立数位分身，过程只是坐在光线充足的房间里，让手机镜头对著脸，朗读两位数数字，再向右看、向左转头。

他生成了两段各 10 秒的影片：一段是在 San Francisco 的 Dolores Park 为恐龙唱生日歌，另一段是在 Golden Gate Bridge 下冲浪。画面虽有结巴、跳切、杯子蛋糕凭空出现、服装不合逻辑等瑕疵，但背景地景、嘴型、脸部轮廓与动作都相当逼真，甚至比一般回看语音信箱或周末影片更像真实本人。

文章同时指出，Google 只允许成人替自己制作 avatar，不像 OpenAI 早先曾让使用者决定是否让他人用自己的样貌生成影片。Google DeepMind 的 Nicole Brichtova 表示，团队试图在不阻挡无害内容的前提下防止伤害；作者则认为这个数位复制体不只是更美化的自己，而是一个可随时到场、几乎无缝的 Reece，带来强烈不安。

Reece Rogers tested Google Gemini app’s avatars feature on May 21, 2026, after paying $20 per month for Google AI Pro; his usage cap reset every five hours, and he quickly maxed it out. It took about five minutes to create his digital double by sitting in a well-lit room, facing his phone camera, reading two-digit numbers, then looking right and swiveling left.

He generated two 10-second clips: one of him singing happy birthday to a dinosaur at Dolores Park in San Francisco, and another of him surfing beneath the Golden Gate Bridge. The results were imperfect—stutters, jump cuts, a cupcake appearing from nowhere, and illogical outfits—but the setting, mouth movements, facial structure, and gestures were strikingly lifelike.

The article notes that Google only allows adult users to make videos with their own avatar, unlike OpenAI’s earlier approach that let users decide whether others could generate videos with their likeness. Nicole Brichtova of Google DeepMind says the team tries to prevent harm without blocking benign uses; the author concludes the clone felt less like an idealized self and more like a seamless, unsettling Reece who can be anywhere and do anything.