Google Gemini Omni Creates Deepfake Toy Video
SkimNews Take
The seamless creation of a "deepfake vacation" for a plush toy highlights how readily advanced AI models can be applied to generate personalized, emotionally resonant media, even for trivial purposes.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google released Omni Flash, the first model in its Gemini Omni family, now available in the AI video generation and editing platform Flow.
- Omni can generate video from an uploaded clip and a text prompt, and Google claims it incorporates more real‑world knowledge and keeps characters consistent.
- Buddy the author's stuffed deer was animated by Omni on a rafting adventure, a cruise, and a skydiving clip, but the model sometimes flipped the deer’s orientation mid‑action.
- Omni introduced visual artifacts such as a honey jar that changed into a bottle and added antlers to Buddy that later disappeared after prompting.
- Omni costs 15‑40 credits per generated clip and 40 credits per edit, quickly depleting a $20‑per‑month AI Pro plan that provides 1,000 credits.
- Omni can add AI‑generated content to real footage, producing deepfakes of the author eating spaghetti, sitting on a plane, and standing before the Eiffel Tower, with minor glitches like a repeated background woman and unnatural fork sounds.
Why it matters: Google's Gemini Omni lets users create realistic video deepfakes with simple prompts, but each clip costs 15‑40 credits and edits cost 40 credits, quickly draining the 1,000‑credit monthly AI Pro allotment. The high cost and visual glitches—orientation flips, disappearing antlers—limit practical use for casual creators.
Ask SkimNews
