Fal's H3 Max Live generates video faster than real-time playback

[AINews] Fal’s H3 Max Live breaks the infinite videogen barrier

Fal post-trained Minimax's H3 model, then ran it on their own inference engine at 35x the official endpoint speed. The result is H3 Max Live: video generates faster than you can watch it. Ethan Mollick first flagged the milestone; Fal employees turned it into an infinite Twitch stream, and Pieter Levels built a similar interactive feed. Twitch and YouTube quickly banned Fal's streams, so Fal launched fal.live. Quality is still slop, but this is the floor for real-time video generation. The post doesn't disclose pricing or exact latency numbers.

Why it matters: Fal pushed Minimax H3 to 35x speed, making video generation faster than playback for the first time — a real engineering milestone. Not scoring higher because we only have Fal's own numbers and one Twitch stream; no third-party reproduction or broader quality comparisons yet.