Google Turns Photos Into Videos With AI—Here's How It Works

Google has added a new feature to its Gemini app that can turn a still photograph into an eight-second video, complete with sound. The feature is now available to paid subscribers in more than 150 countries. Those paying for Google AI Pro can make up to three videos a day, while Ultra subscribers get five per day.

This is Google's latest step in bringing AI video-making tools to regular users. Every video the system creates gets a visible watermark, plus an invisible digital signature called SynthID to mark it as AI-generated. The watermark is there to help people know the video came from artificial intelligence, not a real camera.

How Google Built This Technology

Google didn't build this feature overnight. The company has been working on the underlying technology for years, studying how to process images, video, and sound all at the same time in an efficient way.

Two earlier research projects laid the groundwork. One, called the Multimodal Bottleneck Transformer (MBT), found a way to cut down the computing power needed to handle images and video together by about half. Another, called AVFormer, worked on combining video and sound in ways that machines could understand.

Think of it this way: making a video from a photo is like asking a computer to imagine what happens next in a scene. The computer has to create hundreds of new frames in sequence and match them with audio. That takes a lot of processing power. Google's earlier research made that process faster and cheaper, which is why the company can now offer it to millions of people.

Why There Are Daily Limits

The reason Google limits you to three or five videos per day depends on the kind of computing resources video generation requires. Creating an eight-second video is far more demanding than writing text or generating a single image. The computer has to process thousands of frames, and then layer in audio on top of that. Both take enormous amounts of energy and computing time.

The daily limits tell us that Google is managing costs by controlling how many videos people can make, rather than trying to let everyone make as many as they want. As this technology gets cheaper to run over time, those limits might go away.

The Watermark Question

The visible watermark and the invisible SynthID signature serve two different purposes. The visible watermark is straightforward—anyone looking at the video can see right away that it came from AI. The invisible signature is more technical: it's baked into the video as it's being created, which makes it much harder to remove than a watermark added afterward.

The bigger picture here is that Google has learned from earlier mistakes with AI tools. When the company (and others) first released image-generating AI to the public, there were real concerns about misinformation and people not knowing what was real. Google is trying to get ahead of those same issues with video by making it clear from the start: this is not a recording of something that actually happened.

Whether this two-part approach actually stops people from using AI videos to mislead others is still an open question. Determined bad actors may find ways around the watermarks. But for most people, the visible marker should make clear what they're looking at.

What This Means Going Forward

Google's approach here—offering the feature through a paid subscription with daily limits—gives other tech companies a template for how to roll out video generation carefully. The company is taking time to let people get used to AI video, rather than flooding the world with unlimited free tools right away.

This also shows something broader about the tech industry: the companies with the biggest research teams and the most powerful computers tend to stay ahead. Google spent years on the underlying research, then years more building the actual product. That kind of sustained effort gives them an advantage that smaller competitors will struggle to match.

As video generation tools get cheaper and more companies offer them, these daily limits will probably relax. We've seen this pattern before with image generation—what started as restricted access has become much more open. But for now, Google is moving deliberately, which seems like a reasonable choice given how new and powerful these tools are.