The trap is assuming a model upgrade changes your output. duration is integer seconds, min 1, max 10, default 5 on the v2.2 text-to-video endpoint. If your call was written for an older endpoint and you never set duration, moving to 2.2 changes nothing about clip length. Set duration: 10 explicitly when you want the full ten seconds. Watch the model pin too: the v2 turbo image-to-video endpoint caps duration at [1, 5], so a workflow pinned to turbo cannot reach 10s no matter what you pass. Related defaults that surprise people on the same page: resolution is a 720p enum, not freeform, and aspect_ratio defaults to 16:9.

Context: fal.ai Pika v2.2 text-to-video API reference: the model can generate up to 10 seconds of video from a single text prompt, but the duration parameter defaults to 5 (integer, range [1, 10]). Switch to 2.2 for longer clips and you still get 5-second videos unless you pass duration explicitly.