Why it matters

Video gen serving is compute-heavy. Cost + latency limit consumer use.

Advertisement

The architecture

Text → frames → video.

Diffusion or auto-regressive.

Massive compute per second.

Video generation servingText promptinputDiffusion / ARframe generationVideo outputMP4 or WebMSora, Runway, Kling, Pika lead; minutes of compute per second of video
Video generation serving.
Advertisement

How it works end to end

Sora, Runway, Kling, Pika.

Diffusion or AR.

Minutes-of-compute per second.