Kling 4.0 has the most interesting billing model in the current AI video lineup: the rate you pay depends on whether your request includes video input. This guide makes the rules easy to apply.
Two rate tiers exist per resolution. Without video input — pure text, keyframes, reference images, voices or image subjects — you pay the standard per-second rate on the output duration. With video input — reference videos or video subjects attached — the per-second rate drops, but the input video duration is billed as well.
The video-input formula: when total input duration is less than or equal to output duration, you are billed for twice the output length. When input is longer, you are billed for input plus output. A 10-second 720p output built from a 5-second reference video bills as 20 seconds; the same output built from a 15-second reference bills as 25 seconds.
Because the with-video rate is roughly half the no-video rate, attaching a short reference video can actually cost about the same as generating without one — while giving the model a concrete motion target. Long reference videos are where costs grow: trim them to the essential seconds.
Duration is flexible from 3 to 30 seconds, and resolution is 720p or 1080p with the 1080p rate double the 720p rate in both tiers. As with every per-second model, generating short drafts first and re-running the winner at full length is the cheapest way to iterate.
The Kling 4.0 studio shows a live cost estimate that already accounts for your attached videos' durations — the number you see is the number you pay.