Video
Kling
Kuaishou’s Kling models: Kling 4.0 makes 3 to 30 second clips with sound from text, frames or up to 10 reference images; Kling 3.0 makes 3 to 15 second clips, with sound you can turn off to pay less.
Which to pick
- Kling 4.0 for the best Kling: clips up to 30 seconds in one take, up to 10 reference images to keep people, products and places the same, and 21:9 for widescreen. Every clip has sound.
- Kling 3.0 for a lower price, silent clips that cost less again, and a
negative_prompt.
Prices
A clip costs its length times the rate: a 5 second 720p Kling 4.0 clip is $0.75. Frames, reference images and the shape don’t change the price. It’s charged only when the video is delivered.
Kling 4.0
- A prompt of up to 8,000 characters. There’s no
negative_prompt: say what to avoid in the prompt itself. - A
first_frameto animate, and optionally alast_frameto land on; the video takes the first frame’s shape. - Or up to 10
reference_imagesof people, products, places or a style, mentioned in the prompt as@Image1,@Image2and so on, in the order you sent them. Frames and reference images don’t go together. - Lengths from 3 to 30 seconds, in whole seconds.
- 16:9, 9:16, 1:1 and 21:9. With reference images, adaptive lets the prompt or the images set the shape.
- Sound always comes with the picture: dialogue, ambient sound, music and effects, as the prompt describes them.
- 720p and 1080p.
{ "model": "kling-4.0", "reference_images": [ "https://example.com/hiker.jpg", "https://example.com/coffee-bag.png" ], "prompt": "The hiker from @Image1 walks a ridge at sunrise, then lifts the coffee bag from @Image2 to the camera and says: \"Fresh roast, every morning.\" Keep his face and jacket from @Image1 and the bag's label from @Image2. Wind and birdsong.", "seconds": 8, "resolution": "1080p", "aspect_ratio": "16:9"}Kling 3.0
- A prompt of up to 2,500 characters, and optionally a
negative_prompt(up to 2,500) of things to keep out. - A
first_frameto animate, and optionally alast_frameto land on. Kling 3.0 doesn’t take reference images. - Lengths from 3 to 15 seconds, in whole seconds.
- 16:9, 9:16 and 1:1. From a first frame, the video always takes the image’s shape.
- Sound is on by default; set
generate_audiotofalsefor a silent clip at the lower price. - 720p and 1080p: the higher resolution is a sharper, more detailed render.
{ "model": "kling-3.0", "first_frame": "https://example.com/bakery.jpg", "last_frame": "https://example.com/street.jpg", "prompt": "The camera glides back from the bakery counter, out through the door and onto the street.", "negative_prompt": "text, watermark, blur", "seconds": 5, "resolution": "720p", "generate_audio": false}Prompting
- Lead with the subject and what it does, then the camera: “a cyclist crests the hill at sunrise, low tracking shot, then the camera cranes up”.
- Name camera moves plainly: push in, pull back, pan left, orbit, crane up, handheld.
- With reference images, say what each is for: “keep the face, hair and clothes from @Image1”, “the kitchen from @Image2, but at night”. Without that, the model guesses.
- Write dialogue in quotes next to the character who says it, and name the sounds and music you want.
- For long Kling 4.0 clips, describe the events in order, beat by beat.
- From a first frame, describe the motion rather than what’s already in the picture.
Good to know
- Most clips are ready in 2 to 4 minutes; long and 1080p clips take longer.
- Prompts or images that may break the content rules can be declined (
content_policy); declined videos are free. - Videos are kept for 30 days; download the ones you want to keep.
Full API details: Video generation. Try it in the dashboard’s Playground, which shows the code for each video.
Questions, or something missing? Ask our support team and we’ll answer by email, usually within a minute.