What if you could input a single still image and make an entire video? While that might sound like something far-fetched, it’s actually something you can do with Google’s latest AI endeavor, the Vlogger AI video generator.
Google researchers unveiled Vlogger this week, showcasing the AI model’s capabilities in multiple videos and a paper that details the AI’s capabilities. The tool allows the user to input a reference image and then feed various commands to the AI to help determine what the video will look like. It’s a bit of a different approach to video generation compared to OpenAI’s Sora, but it could still have a lot of uses.
There are, obviously, some concerns around these capabilities. For starters, what is to stop anyone from simply inputting an image of any random person and then creating AI videos of them? Google’s Vlogger AI video generator will let you do just that. It’ll also let you match up the mouth of a person to an audio track, even if it’s in another language, so you could technically create Spanish content using someone who doesn’t speak Spanish.
Based on the information available in the paper, and on the project’s website, it appears that you may also be able to command the AI generator to move the persona’s body around. It’s possible I’m misunderstanding the notes provided in the project information, but if that is the case, it certainly offers quite a lot of control over the video.
Of course, the tech is far from perfect, and there are still those tell-tale signs that Vlogger is generating these videos using AI. In some examples, the mouth movements feel very unnatural, which gives away the use of AI to create the video. But that’s the thing with AI like this, it doesn’t have to be perfect to be useful. And, this is the worst that Vlogger is ever going to be. As time goes on, and Google feeds more material into the model, it’s only going to get better and better.
I’m not sure about you, but that’s both terrifying and exciting. For now, at least, Vlogger’s AI video generator isn’t perfect enough to prove too useful to deep fakes or misinformation campaigns. But, if we don’t get a hold on how to handle AI going forward, it could very well have a place in those fields once Google improves it even more.
