Google Gemini 1.5 Pro Flags Sora-Generated Videos as Non-authentic

2024-02-18

"In general, this video is visually appealing, but there are many inconsistencies that indicate it is not a real-life scene." "Google's Gemini 1.5 Pro questions the authenticity of the video generated by OpenAI's Sora, labeling it as false and pointing out significant inconsistencies." "Previously, tech giants Google and OpenAI announced their latest developments - Gemini 1.5 Pro and Sora, respectively. The strategic timing of Sora's release sparked speculation that it was intentionally diverting attention from Google's Gemini 1.5." "In response, Google shared a detailed analysis on X, criticizing the video produced by Sora." "Gemini 1.5 Pro dissects a scene of a snow-covered street adorned with cherry blossoms in Japan. The analysis points out several inconsistencies that raise doubts about the video's authenticity." "According to Gemini 1.5 Pro, the simultaneous presence of heavy snowfall and blooming cherry blossoms caught people's attention because cherry blossoms typically bloom in spring without snow." "Further examination revealed that the snowfall in the video was uniform and unnatural, whereas real snowfall is irregular. Additionally, despite the heavy snowfall, the characters in the video were not wearing any winter clothing." "In conclusion, Gemini 1.5 Pro states, 'Overall, the video is visually appealing, but the inconsistencies indicate that it is not a scene from real life.'" "Sora is OpenAI's new cutting-edge text-to-video tool that can create videos up to 60 seconds long, featuring highly detailed scenes, complex camera movements, and characters with vivid emotions. Many also refer to it as ChatGPT in video generation." "Google's Gemini 1.5, on the other hand, boasts an astonishing 1M token context window, surpassing not only GPT-4 Turbo's 128K but also Anthropic Claude 2.1's 200K. It can handle a large amount of information at once, including one-hour videos, 11-hour audios, and codebases exceeding 30,000 lines or 700,000 words."