Two mystery AI video models — codenamed “Polaris” and “Vega” — just started showing up in blind testing on the Artificial Analysis arena, and nobody has confirmed who made them.
There’s no official attribution yet, but between the timing and the quality, I’d be surprised if at least one of these isn’t Meta’s long-promised video model. Could be Kling 4? Could be a dark horse?
In this one I break down every output we’ve got — multi-shot character consistency, physics, on-screen text, native audio, and one extremely uncensored baby Batman — and tell you which of the two I’d take head-to-head.
Plus: StillsLab, a free shot-reference library built for exactly the way AI filmmakers work — color palettes with hex codes, camera and lens data on every still, and a visual search I haven’t seen anywhere else: feed it your AI generations and it finds real film shots in the same visual family.