Google’s Gemini ad meets reality in a deepfaked stuffed deer test

A stuffed deer named Buddy was supposed to live only in Google’s holiday fantasy, but one parent took the idea further and made the plush animal look as if it had gone on vacation. The result was more than a novelty. It showed how quickly “anything-to-anything” video generation can turn a playful marketing hook into something that looks strikingly real.

The experiment started with a simple question: could the scenes in Google’s Gemini ad be recreated well enough to feel believable? The answer was yes, at least to the point that the videos of Buddy the deer on his adventures were never shown to the parent’s four-year-old. That restraint matters. A tool that can produce charming family keepsakes can also produce convincing fabrications with almost no friction, which is exactly why the line between creative play and manipulation is getting thinner.

Google had already built the concept into its holiday Gemini campaign, which featured parents using Gemini to help locate or replace a child’s lost stuffed animal. Buddy the deer became the emblem of that pitch. By late 2025, Google had expanded the idea into a broader holiday ad series, using stuffed-animal storytelling to show off Gemini’s consumer appeal. The message was clear: generative AI could help parents turn loss into a small, emotional rescue story.

Now Google is pushing the same idea much further. On May 19, Google announced Gemini Omni, a new multimodal model family that it says can create anything from any input, starting with video. Google says Omni can take images, audio, video and text as input, then generate high-quality video grounded in Gemini’s real-world knowledge. The first model in the family, Gemini Omni Flash, is rolling out to the Gemini app, Google Flow and YouTube Shorts.

That rollout points to the commercial prize and the policy problem. For creators, the ability to edit video through conversation and mix media inputs could compress production timelines and lower the cost of making polished clips. For everyone else, the same capability makes deceptive video easier to produce, easier to scale and harder to flag. The stuffed deer test shows the technology is already good enough to blur the boundary. What it does not solve is how viewers will know when a scene is a genuine moment, a branded story or a synthetic one built to look real.