Google’s Gemini Omni can turn your videos into convincing deepfakes

In short: A hands-on test shows Google’s new Gemini Omni can edit and generate videos that look very real, including deepfakes made from a simple selfie clip.

What happened

Google has introduced a new set of AI models called Gemini Omni, which it says can eventually turn one kind of input into another, like text into video or video into something new. For now, the first released model, Omni Flash, focuses on making and editing video.

In a test for The Verge, reviewer Allison Johnson used Omni Flash inside Google’s Flow platform. She created short videos of a stuffed deer on vacation, using a mix of prompts (typed instructions) and uploaded video. Some clips matched her instructions better than older Google video models, but there were still obvious glitches, like objects changing shape and a character suddenly flipping position mid-scene.

The more striking test was a deepfake. A deepfake is a video that makes it look like a real person did something they did not do (like a very convincing impersonation). Starting from a neutral selfie video, Johnson asked Omni to generate clips of her eating spaghetti, sitting on an airplane, and posing in front of the Eiffel Tower. She said the results were convincing enough that her husband believed the pasta clip was real, with only small “tells” like repeated background people or slightly artificial sound.

The tool also has a cost. Video generation uses credits, and Johnson said her $20 per month plan included 1,000 credits, which dropped quickly after making around 20 clips with some edits.

Why it matters

Tools like this make it easier for everyday people to create realistic fake videos, for fun or for misleading posts online. It also means viewers may need to be more cautious about trusting short videos as proof of where someone was or what they did.

Source: The Verge AI