I tried China's hot new vibe coding app. One feature is light-years ahead of ChatGPT. (www.businessinsider.com)

China’s new “vibe-coding” app LingGuang, built by Ant Group and launched Nov. 18, exploded to more than 2 million downloads and briefly crashed its “flash app” feature from demand. Its standout is an “AGI camera” that analyzes scenes in real time (no photo upload) to identify people and products, surface item attributes, answer voice queries, and even generate short edited videos or animated clips directly from captured images. The app also creates mini-apps in about a minute via a flash-generator workflow and guides users with tappable prompt bubbles, delivering a seamless, AR-like experience that felt faster and more integrated than the author’s ChatGPT comparison. ChatGPT matched the analytical depth when given photos but requires manual uploads, saves images in chat, and outputs code rather than instant runnable apps or on-device-style video edits. For the AI/ML community this signals notable progress in multimodal, low-latency inference and end-to-end UX: real-time scene understanding plus immediate media synthesis represents a jump in continuous capture → edit → generate pipelines. Trade-offs are clear — LingGuang doesn’t persist session data (no saved photos/responses) and doesn’t expose generated app code, limiting reproducibility and developer customization. There are also privacy and model-origin questions (accurate person/product recognition, web-sourced comparisons) that will matter as these real-time, on-device-style capabilities scale. Overall, LingGuang advances immediacy and interaction design for multimodal AI while leaving deeper developer control and auditability to tools like ChatGPT.