Models
Which model strings the API accepts today, and how the kling-4-0 upgrade will work when it ships.
Available now
| Model string | Status | Notes |
|---|---|---|
kling-v2-6 | Default | The current production Kling model on our pipeline. Used when model is omitted. |
kling-v2-6 supports everything documented in Generate a Video: std/pro modes, 5s/10s durations, three aspect ratios, native audio (pro), negative prompts, and up to two reference images (end-frame control in pro).
Kling 4.0 at launch
Kuaishou has not announced an official Kling 4.0 release date — we track the signals in the release-date post. When it ships on our platform, it arrives as a new model string, and upgrading is a one-field change:
{
- "model": "kling-v2-6",
+ "model": "kling-4-0",
"prompt": "…",
"parameters": { … }
}Everything else — auth, endpoints, job lifecycle, response shape — stays the same. That's the point of the job-based design: model generations change, the integration doesn't.
Requests with a model string the API doesn't recognize fail with 400 INVALID_INPUT, so you can deploy the new string the moment we announce availability and not before.
Which mode should you use?
std— cheaper and faster; text-to-video and single-reference work. Good for iteration: find your prompt in std, then re-render the keeper in pro.pro— higher render quality, plus the pro-only features: native audio and two-image end-frame control.
For prompt technique — shot lists, camera language, film-stock grading — see the multi-shot prompt guide.