Is Google’s Gemini Demo a Staged Illusion? The Truth Behind the AI Showcase

Google’s newest Gemini AI model generated mixed reactions after its debut, but the most eye‑catching part of the company’s demo was almost entirely fabricated, causing users to doubt Google’s integrity.

The video titled “Interacting with Gemini: Experiencing Multimodal AI” amassed a million views and showcased a sketch of a duck evolving from a simple line, surprise at a blue toy duck, voice queries about the toy, a cup‑changing game, shadow‑gesture tracking, planetary sketch rearrangements, and more.

The video is not genuine. Bloomberg’s Parmy Olsen reported that the team captured footage, then fed static image frames to Gemini with textual prompts, creating the illusion of real interaction.

Although Gemini can generate the shown responses, the demo edited out latency and lengthy answers, presenting an idealized, shortened interaction rather than a faithful real‑time performance.

Google’s blog post ( https://developers.googleblog.com/2023/12/how-its-made-gemini-multimodal-prompting.html ) reveals the actual prompts and responses used.

Specific examples—such as the hand‑gesture rock‑paper‑scissors game, the planet‑ordering question, and the paper‑ball‑under‑cup tracking—required carefully engineered prompts and multiple image frames, not spontaneous perception.

Thus, the video is a stylized impression rather than a true demonstration, undermining trust in Google’s claims about Gemini’s capabilities.

Later, DeepMind VP of Research Oriol Vinyals posted additional details on how the video was produced, emphasizing that it was meant to inspire developers, not to serve as an unedited showcase.