MOCKUP / EXAMPLE — new /apps/{provider}/{model} template applied to klingai/video-v3. Pricing & runtime values are placeholders, fill from real data. Spec: model-page-template-spec.md
apps / video / klingai/video-v3
new text + image → video made in China up to 4K

Kling V3

Run Kling's latest and most capable video model online or via API. Native 4K, multi-shot generation, 3-15s clips with synchronized audio, motion and element control.

run in browser → run via API
// try it now

Run Kling V3 in your browser

Same interface you get inside the app. Drop in a prompt or image, hit run. Sign in to generate.

form { } json >_ api
click to upload · or drag and drop
example output
▶
prompt: "neon-lit Tokyo alley, push-in" · 5s · 4K
Sign in to run Kling V3

Free account, pay only per second of video. Your inputs carry over.

run this in browser →

runs on inference.sh · no GPU setup · cancel anytime

// overview

What Kling V3 does

Kling V3 is the third-generation video model from Kling AI, built by the Chinese company Kuaishou. You give it a text prompt, a starting image, or both, and it generates a short video clip — now at native 4K with synchronized audio, where earlier versions topped out at 1080p and silent output.

The headline upgrades over V2 are multi-shot generation (a single prompt can produce a sequence of camera angles instead of one continuous take), finer motion and element control, and flexible 3 to 15 second durations billed per second so you only pay for the length you need. It is a strong fit for cinematic b-roll, social and UGC ads, and product or character shots where camera movement matters.

Because it is a closed, API-only model, you cannot self-host it — you run it through a provider. On inference.sh you call it with the same API key and billing as every other model, with no per-vendor signup.

// example outputs

What it generates

Real generations. Hover any clip in the app to see the exact prompt and settings.

▶ 4K · 5s
text-to-video

"Aerial drone shot over turquoise coastline at golden hour, slow orbit."

▶ 4K · 8s
image-to-video

Product still animated into a 360° rotating studio shot with soft light.

▶ 4K · 10s
multi-shot

"Character walks through a market" rendered as three cut camera angles.

// inputs

Inputs, explained

The full schema, in plain English. Defaults are sensible — change these when you need to.

prompt *
string · max 2500
What you want to see. Describe subject, motion, and camera move ("slow push-in", "orbit"). The clearer the camera direction, the more control you get.
image
string (file)
Optional start frame for image-to-video. jpg, jpeg or png, max 10mb, min 300px. Leave empty for pure text-to-video.
end_image
string (file)
Optional end frame. Requires image to be set as the start frame. The model interpolates motion between the two.
mode
string · std / pro
"pro" for higher quality, "std" for faster and cheaper drafts. Default pro.
duration
integer · 3-15
Clip length in seconds. You are billed per second, so this is the main cost lever. Default 5.
aspect_ratio
string · default 16:9
16:9 for landscape, 9:16 for vertical/social, plus 1:1, 4:3, 3:4, 3:2, 2:3 and 21:9.
sound
boolean · default true
Generate synchronized audio with the video. Turn off if you are scoring the clip yourself.
negative_prompt
string · max 2500
What to avoid in the video (artifacts, objects, styles). Optional.
// api reference

Run via API

One key, one bill, every model. This is the standard inference.sh API reference for klingai/video-v3.

about

Kling V3.0 — latest and most capable video generation model. Native 4K output, multi-shot generation, flexible 3-15s duration billed per second, element control, motion control, and synchronized audio.

1. calling the api

Install the client, set your API key, then submit a request and wait for the result — or stream live progress.

pythonjavascriptcurl
# install the client pip install inferencesh # set your api key — get it from app.inference.sh/settings/keys export INFERENCE_API_KEY="inf_your_key" from inferencesh import inference client = inference() # run and get result result = client.run({ "app": "klingai/video-v3", "input": { "prompt": "neon-lit Tokyo alley at night, slow push-in", "duration": 5, "mode": "pro" } }) print(result["output"]) # or stream live progress updates for update in client.run({ "app": "klingai/video-v3", "input": { "prompt": "..." } }, stream=True): if update.get("progress"): print(f"progress: {update['progress']}%") if update.get("output"): print(f"output: {update['output']}")

2. authentication

The API uses API keys for authentication. Set INFERENCE_API_KEY as an environment variable. See the authentication docs for detailed setup.

3. files

File inputs are handled automatically by the SDK. The Python SDK detects local file paths and uploads them; URLs are passed through as-is. You can pass local paths, URLs, or base64 data.

python
# local file paths are automatically uploaded result = client.run({ "app": "klingai/video-v3", "input": { "image": "/path/to/local/image.png", # detected & uploaded "prompt": "animate this still into a slow orbit" } })

4. webhooks

Get notified when a task completes by providing a webhook URL. When the task reaches a terminal state (completed, failed, or cancelled), a POST request is sent to your URL with the task result.

pythonpayload
result = client.run({ "app": "klingai/video-v3", "input": { }, "webhook": "https://your-server.com/webhook" }, wait=False) # your endpoint receives a JSON POST: # { # "id": "task_abc123", # "status": 9, # 9=completed, 10=failed, 11=cancelled # "output": { ... }, # "error": "", # "created_at": "2024-01-15T10:30:00Z", # "updated_at": "2024-01-15T10:30:05Z" # }

5. schema

input

prompt *
string
Text prompt describing the video content. Max 2500 chars.
image
string (file)
Start frame image for image-to-video. Formats: jpg, jpeg, png. Max 10mb, min 300px.
end_image
string (file)
End frame image. Requires image to be set as start frame.
mode
string · default "pro"
Generation mode. 'pro' for higher quality, 'std' for faster/cheaper. Options: std, pro.
sound
boolean · default true
Generate synchronized audio with the video.
aspect_ratio
string · default "16:9"
Video aspect ratio. Options: 16:9, 9:16, 1:1, 4:3, 3:4, 3:2, 2:3, 21:9.
duration
integer · default 5
Video duration in seconds (3-15). Billed per second. Min 3, max 15.
negative_prompt
string
What to avoid in the video. Max 2500 chars.

output

video *
string (file)
The generated video file.
// related tools

Similar & companion tools

Other video models

Google Veo 3
US · text-to-video, audio
Runway Gen-4
US · video gen + editing
Hailuo / MiniMax
China · fast i2v

More from Kling AI

Kling V2.5
prior version · 1080p
Kling Lip Sync
re-sync video to audio
Kling Motion Brush
paint motion paths

Pairs well with

ElevenLabs Voice Design
make the voice track →
HeyGen Lip Sync
re-dub the result →
Seedream 5
generate the start image →
// faq

Common questions

How much does Kling V3 cost?
It is billed per second of generated video, so a 5-second clip costs roughly five times a 1-second one. 4K runs cost more than 1080p. On inference.sh there is no subscription — you pay per use on the same bill as every other model. (Plug exact rate here.)
Can I use Kling V3 videos commercially?
Yes. Outputs can be used commercially. As with any generative model, you are responsible for the inputs you provide (don't upload copyrighted source images you don't have rights to).
Is Kling V3 a Chinese model?
Yes. Kling AI is developed by Kuaishou, a company based in China. If data residency matters for your use case, note that requests are processed by the provider. inference.sh routes the call but the model itself is Kling's.
What's new in V3 vs Kling V2?
Native 4K output (up from 1080p), synchronized audio, multi-shot generation from a single prompt, finer motion and element control, and flexible 3-15s duration billed per second.
Can I run Kling V3 locally?
No. It is a closed, API-only model with no public weights, so it must run through a provider. inference.sh gives you API and browser access without a separate Kling account.

Ready to run Kling V3?

Sign in and generate your first 4K clip in a click. Pay per second, no setup.

run in browser →