How to Spot AI-Generated Fake Videos: Tips from Storyful

Understanding AI Video Generation

• 💡 OpenAI's new AI video generator, Sora 2, is capable of turning text prompts into convincing videos.
• 🚀 A key development allows users to insert themselves and their voice and likeness into generated videos.
• ⚠️ This advanced technology raises concerns about the authenticity of video content for audiences and newsrooms.

Clues to Identify Fake Videos

• 🔍 Examine text within videos; AI often struggles to render characters correctly, leading to garbled words (e.g., 'gratics' instead of 'graphics').
• ⏱️ Analyze timestamps; surveillance footage timestamps that appear nonsensical or have garbled characters are a strong indicator of manipulation.
• 🤸 Look for physics-defying activity; unusual movements of objects without apparent cause, like a box moving on a shelf, can signal AI generation.

Verifying Video Authenticity

• 📰 Storyful, an expert in content verification, emphasizes the importance of checking the source of a video.
• 🌐 Corroborating information from authoritative sources and official statements (e.g., from Target regarding a shoplifting event) is crucial for validation.
• ✅ Performing these rudimentary checks can quickly expose manipulated or fake video content.