Can we really tell what’s made by Sora 2? 11 tips to help spot AI-generated video (www.techradar.com)

OpenAI’s Sora 2 is an upgraded text-to-video model that generates short, highly realistic clips — good enough that even experienced researchers and journalists can be fooled. Unlike earlier video AIs, Sora 2 largely fixes obvious artifacts (bad hands, wonky physics, blurred faces), pushing generated footage into a new realism threshold. That raises the stakes: what starts as harmless novelty could scale to convincing deepfakes of politicians, staged arrests, or manipulated personal footage, making media literacy and verification practices essential for the AI/ML community and the public. The article offers 11 practical detection tips that cluster around a few technical tells: inspect backgrounds for impossible geometry or inconsistent extras; watch for physics and lighting errors (mismatched shadows, reflections, objects popping in/out); notice unnatural motion (overly smooth or oddly timed blinks, hair/fabric moving without cause); and look for compression artifacts, smudges or overly clean patches that betray synthesis. Also be wary of emotional manipulation, moving watermarks that can be faked or removed, and single-source viral posts—scrutinize the account, comments, metadata, and corroboration from reputable outlets. The key takeaway: Sora 2 raises the bar for realism, so detection now requires deliberate slow-down, cross-checking, and more robust verification workflows as a core part of digital trust.