Kling 3 seven-second clip: $1.47 with sound, $0.98 silent on Sume

A seven-second Kling 3 clip (kling-3) costs $1.47 on Sume when audio is on, which is the default, and $0.98 when you send generate_audio: false. The rate does not change between 720p and 1080p, so the only knobs that move the price are length and the audio flag.

This post is arithmetic from the repo's rate table, so you can reproduce it. Sume prices come from the rate tables in the repo (provider list per second) times the 1.25 house margin, rounded up to whole cents, as the catalog's billable formula states.

The two rates

The Kling row in the Video Router catalog stores two provider list rates: $0.112 per second with audio off and $0.168 per second with audio on. Sume sells at list times 1.25, which gives $0.14 and $0.21 per second.

Audio-on is the default because the /v1/videos docs say generate_audio defaults to the audio capability of the model, and Kling 3 advertises audio. If you only need a silent clip for a timeline that carries its own music, send false explicitly and save a third of the price.

Where 7 seconds sits in the range

Kling 3 on Sume accepts 4 to 15 seconds, 720p or 1080p, and 16:9, 9:16 or 1:1. Seven seconds is a common length for a product reveal plus a hold, because it is long enough for a camera move and short enough that one job covers a social slot.

Submit it

The request below pins the model, asks for a vertical clip at 1080p and turns audio off. Sume reserves the billable amount when you submit and usage.cost on the finished job shows what was captured.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kling-7s-001" \
  -d '{"model":"kling-3","prompt":"A slow push-in on a ceramic mug, steam rising","duration":7,"resolution":"1080p","aspect_ratio":"9:16","generate_audio":false}'

Two limits to know

Kling 3 on Sume takes no input_references: the catalog row says no reference image, video or audio URLs. A first frame through frame_images works, and so does a last frame. If you need references, pick Wan 3.0 or Seedance 2.5 instead.

Always read GET /v1/videos/models before a bulk run. It is the source of truth for durations and ratios, and it returns a 400 for values the model does not advertise rather than silently changing your request.

• Length: 4 to 15 seconds, whole seconds only.
• Resolutions: 720p and 1080p, same per-second price.
• Audio: default on at $0.21 per second, off at $0.14 per second.