Fal Unveils Live AI Video Steered by Viewer Prompts

Fal says its H3 Max model's faster-than-real-time generation now powers fal.live, a product that lets viewers reshape an AI video stream in real time using prompts.

Fal has introduced a new use case for its H3 Max generation model that moves beyond producing individual video clips and into continuous, interactive streaming. According to the company, the model's ability to generate frames faster than real time makes it possible to run a live AI video feed that responds to audience input on the fly rather than being rendered once and played back. The product built on this capability is called fal.live, which fal describes as letting viewers steer an ongoing AI-generated stream by typing prompts. Rather than waiting for a new clip to be generated and swapped in, the stream reportedly adjusts within seconds and keeps running without visibly cutting or restarting, so the changes feel like they are happening inside a single continuous broadcast rather than being stitched together after the fact. Fal also indicated that the underlying system is built on a component it refers to as "World," suggesting the streaming behavior may be tied to broader world-model style generation rather than a narrow one-off video generator. The company did not provide additional technical detail on how that piece functions, but framed it as part of what allows the stream to keep responding to new instructions without breaking continuity. The announcement came directly from fal's own X post introducing the concept, and the claims about fal.live's real-time, prompt-driven behavior are self-reported by the company rather than independently verified. Still, the pitch points to a shift in how generative video tools might be packaged: not just as a way to produce finished clips, but as a live, steerable medium that audiences can influence as it plays out.