Kling 4.0 API

Kling 4.0 (Oct 2026) via WaveSpeed — Kling's newest video model: 720p clips of 3–30 seconds with native stereo audio, text-to-video and image-to-video, at $0.157/s. Billed in Varosity Credits.

$0.157/s on Varosity credits ($0.157/s with your own key). Clips up to 30s with native audio.

Use Kling 4.0 with your agent Try it in the browser

no wavespeedaccount needed · provider's list price, no markup · failed renders never billed

Best for

newest Kling, native audio
single takes up to 30s
21:9 widescreen

Trade-offs

slow: a 5s clip can take about 10 minutes
720p only

How to get the best result

Kling 4.0, Kling's newest model, on Credits today: 720p clips with native stereo audio and single takes of 3 to 30 seconds. Pick instead: kling-3.0 for fast 5 or 10s Kling clips; veo-3.1 for 4-8s lip-sync talking heads; grok-imagine-video for cheap audio clips up to 15s. Pick another model when the result is needed within a few minutes.

Prompting

Name the subject, the action and the setting, then the camera movement ('slow dolly-in', 'handheld tracking shot').
Describe the sound you want (ambience, music, a line of dialogue in quotes); audio is generated with the picture.
For long takes, describe what happens in order (opening, middle, end) so the clip has a structure.
Image-to-video: describe the motion, not the picture; leave aspectRatio unset to follow the image.

Settings on Varosity

durationSec is whole seconds from 3 to 30, billed per second produced at $0.157/s.
Always 720p; aspectRatio 16:9, 9:16, 1:1 or 21:9 (others fall back to 16:9).
With referenceImageUrl the image is the first frame and its framing is kept.
Slow: a 5-second clip took about 11 minutes in testing, and longer clips take longer (a 30-second one may run past an hour). Keep polling get_job (every minute or so is plenty) until it finishes; the finished clip is collected and charged on that poll.

Avoid

Using it when the result is needed within a few minutes.
Asking for 30 seconds when 5 will do: cost and render time scale with every second.
Relying on readable text in the scene; signs come out as made-up lettering.

Example prompt

Slow dolly through a rain-soaked night market. A street cook flips sizzling skewers and calls out: "Two for five!" Steam, neon, rain on canvas awnings.

From wavespeed.ai, checked against what Varosity sends.

Call it from your code

One Varosity key works for every model. Video renders are async — the response has a jobId; poll GET /api/v1/jobs/{jobId} until status is "succeeded", then read outputUrl.

curl -X POST https://varosity.ai/api/v1/video/generate \
  -H "Authorization: Bearer $VAROSITY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "cinematic aerial of a coastal road at golden hour, slow push-in",
    "modelId": "ws-kling-4.0",
    "aspectRatio": "16:9",
    "durationSec": 5
  }'

Or ask your agent

Connect Claude, Cursor, or any MCP client to https://varosity.ai/api/mcp with your key, then say:

"Use Varosity with ws-kling-4.0 to make a 5-second clip of …"

Agent setup guide