YouTube has steadily been baking more and more AI into its foundation. Not only has Google made it possible to converse with Gemini about YouTube videos, but we’ve also seen reports of YouTube using AI to “improve” user videos without their general consent. But now, YouTube appears to be taking a page out of Sora’s book and making it possible to create an AI avatar of yourself, which you can then throw into videos to make yourself part of the scene.
Put your face in YouTube Shorts without actually being there
AI picks up a you-shaped mask
It hasn’t been that long since we waved goodbye to Sora, OpenAI’s AI video app which allowed you to throw your face and voice into videos with easy prompts. While there’s still a bit of debate around why OpenAI gutted Sora, the fact is, the death of Sora left a massive vacuum in the AI video space, and now it looks like Google hopes to fill it using YouTube Shorts.
The idea is to allow YouTube creators to record a live video selfie of themselves speaking and looking at the camera, according to a YouTube help page. This video is then used to create an AI avatar of the person, which can then be added into different videos and scenarios using a prompt to the AI. It’s literally exactly what Sora used to do, but this time powered by Google’s Veo video models.
The feature as already started rolling out to YouTubers outside of Europe that are 18+, and you can check if its available in the main YouTube app or YouTube Create. To get started, find the Remix menu, then select Reimagine > Add me to this scene, or select the Create Video option and choose Make a video with my avatar.
Videos will come with built-in AI watermarksBut it is an imperfect solution for calling out AI
Like many of Google’s AI products, the new Shorts AI avatar system will provide videos that utilize it with a visible watermark as well as digital labels like C2PA and SynthID. While technically great at helping to identify AI videos, the physical watermarks added to videos are never going to be a great mark for accuracy in the social media world, where it’s just as easy to go in and crop out the watermark or remove it through editing.
This is, of course, where the digital labels come into play. But the problem with SynthID and C2PA is they require people to go out of their way to confirm if something is AI or not, and that’s not likely something people are going to be willing to do all that often, especially when they’re just scrolling through YouTube Shorts on their break at work.
Considering how much of a problem deepfakes on YouTube have become, I can’t imagine this is going to minimize that issue in any way. If you are interested in trying it out for yourself, though, it should hit your YouTube sometime within the next few days to weeks.
