kling-4.0Kling 4.0 is the newest video model in Kuaishou's Kling family, and every clip it makes comes with sound generated in the same pass. On APIMODELS it is one model id, kling-4.0 (kling-4-0 and kling-v4 also work), on the unified video endpoint, with two modes chosen from what you send. Reference-to-video covers plain text-to-video and adds up to 10 reference images in image_urls, mentioned in the prompt as @image_1, @image_2 and so on; up to 5 reference videos in video_urls (@video_1 and on), each 2 to 30 seconds and 30 seconds in total; up to 7 named subjects in elements, each either an image subject of 2 to 4 views with the main view first or a single-shot character video of 3 to 8 seconds, mentioned as @Name; and up to 3 named voice clips in voices, also mentioned as @Name. One request takes at most 15 materials and at most 3 video subjects, and keep_original_sound keeps the original audio of the reference videos. Keyframe-to-video takes first_frame_url (image and image_url are accepted too), up to 8 middle frames in keyframe_urls and last_frame_url, mentioned as @first_frame, @keyframe_1 and @last_frame; the total number of frames is capped by the duration, from 4 frames at 3 seconds to 10 frames at 9 seconds or longer, and the clip follows the first frame's aspect ratio. Frames cannot be combined with references, and every @mention has to point at an input in the same request; otherwise the request is rejected and nothing is charged. Both modes take a required prompt of up to 10,000 characters, duration as any whole number of seconds from 3 to 30 (default 5), resolution 720p (default) or 1080p, and aspect_ratio auto, 16:9, 9:16, 1:1 or 21:9. Sound is always generated; generate_audio: false returns a 400. Billing is per billed second. Without video input the billed seconds equal the output length, at $0.157/s for 720p and $0.235/s for 1080p, so a 5-second 720p clip is $0.785 and a 10-second 1080p clip $2.35. With reference videos or video subjects the rate is $0.094/s (720p) or $0.141/s (1080p), and the billed seconds are the output length × 2 when the input video is no longer than the output, or input + output when it is longer: a 10-second 720p clip with a 5-second reference video is 20 billed seconds, $1.88, and with a 15-second reference video 25 billed seconds, $2.35. Failed generations are not charged.
Characters, products, outfits or places. Each one is @image_1, @image_2… in upload order.
MP4 / MOV / WebM, 2-30 seconds each and 30 seconds in total. Each one is @video_1, @video_2… Adding video changes the billing, see the estimate below.
Named subjects (elements, @Name) and voice references are available through the API — see the Kling 4.0 docs.
auto follows your reference images and videos; with a prompt only it is 16:9
Sound is generated with every clip and cannot be turned off.
$0.785 for this clip (720p, $0.157/s × 5s) · only successful generations are billed
Generated video will appear here
Provide URLs and click Generate
Any whole length from 3 to 30 seconds at 720p or 1080p. Every clip comes with generated sound; keep_original_sound keeps the audio of your reference videos instead
Up to 10 images and 5 videos, 15 materials in total. Point at each one in the prompt as @image_1 or @video_1, so the model knows which input plays which part
elements turns 2-4 views or a 3-8 second clip into a subject you call by @Name; voices binds up to 3 reference audio clips to names
A first frame, up to 8 middle frames and a last frame, mentioned as @first_frame, @keyframe_1 and @last_frame; up to 10 frames from 9 seconds up
720p $0.157/s, 1080p $0.235/s. With reference video $0.094/s or $0.141/s per billed second, billed as output × 2 or input + output
Failed generations are not charged, and inputs the model cannot take are rejected with a 400 before a task is created
Kling 4.0 is Kuaishou's flagship video model: one API call turns a prompt, up to 15 reference materials or a set of keyframes into a 3 to 30 second clip at 720p or 1080p, with sound generated in the same pass.
On APIMODELS the Kling 4.0 API is a single model id, kling-4.0, on POST /v1/video/generations, billed per second from $0.157, and a failed Kling 4.0 generation is not charged.
kling-4.0What Kuaishou highlights in its new flagship, and where each point shows up in the Kling 4.0 API.
Tracking shots, orbits, complex camera moves
Kuaishou says Kling 4.0 keeps bodies and objects coherent through fast action and complex camera work. Describe the move in the prompt, or borrow it from a reference video.
Always on, no extra charge
Every Kling 4.0 clip comes with sound generated in the same pass, which Kuaishou describes as two-channel stereo. With reference videos, keep_original_sound keeps their own audio instead.
keep_original_soundLanguages, accents and dialects
Kuaishou reports tighter lip sync and speech in several languages, accents and dialects. Give a Kling 4.0 character its voice with a named clip in voices, or bind voice_url to a subject.
voicesvoice_urlMore expressive characters
Kling 4.0 aims at realistic lighting and surface detail and at characters whose faces and bodies react to the scene. Write the performance into the prompt: who speaks, what they do, how they react.
Signs, labels, packaging
According to Kuaishou, on-screen text, logos and small symbols come out clearer than in earlier Kling versions, which matters for product shots. Put the exact wording in quotes in the Kling 4.0 prompt.
Room for a detailed brief
A Kling 4.0 prompt can run to 10,000 characters, enough to describe a 30-second take beat by beat and to point at every reference with an @mention.
promptFour kinds of reference, each called in the prompt by its own @mention.
Kling 4.0 numbers references by array position, so reordering image_urls changes what @image_2 points at. A mention with no matching input in the same request returns a 400 before anything is charged.
Up to 10 · @image_1 to @image_10
Characters, products, outfits, places, composition or a style frame: the who and what of the Kling 4.0 shot.
image_urlsUp to 5 · 2–30 s each · 30 s in total
Motion, camera movement and pacing for Kling 4.0 to follow, mentioned as @video_1. Any video input switches the request to the video-input price.
video_urlskeep_original_soundUp to 7 · at most 3 video subjects · @Name
An image subject is 2–4 views with the main view first; a video subject is one 3–8 second single-shot character clip. Add a tag, a description of up to 100 characters, and voice_url for characters.
elementsUp to 3 · @Name
Standalone reference audio as {name, audio_url}, used only in this Kling 4.0 request. Names are 1–20 characters, unique regardless of case, and cannot be reserved words such as image_1 or first_frame.
voicesThe route is picked from the fields you send. Frames and references cannot be mixed in one request.
A single Kling 4.0 request renders one continuous take of 3 to 30 seconds. In the keyframe route the duration also sets the frame budget: first, middle and last frames together may not exceed the duration plus one, and never more than 10.
A prompt alone, or a prompt plus any mix of images, videos, subjects and voices, up to 15 materials.
promptrequired, up to 10,000 charactersimage_urlsup to 10video_urlsup to 5, 2–30 s eachelementsup to 7 subjectsvoicesup to 3aspect_ratioauto · 16:9 · 9:16 · 1:1 · 21:9Mention in the prompt@image_1 · @video_1 · @Name
A first frame, up to 8 middle frames and a last frame. The Kling 4.0 clip follows the first frame's aspect ratio.
promptrequired, describe what happens between framesfirst_frame_urlfirst frame (image and image_url also work)keyframe_urlsup to 8, in time orderlast_frame_urllast frameMention in the prompt@first_frame · @keyframe_1 · @last_frame
| Kling 4.0 duration | 3 s | 4 s | 5 s | 6 s | 7 s | 8 s | 9–30 s |
|---|---|---|---|---|---|---|---|
| Max frames (first + middle + last) | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
Fields shared by both routes, with their limits and defaults.
3–30 seconds, integer
Default 5. Each Kling 4.0 request makes one continuous take; extending an existing clip is not supported.
duration720p (default) or 1080p
Kuaishou has announced 4K and 10-bit HDR for Kling, but neither is available for Kling 4.0 yet.
resolutionauto · 16:9 · 9:16 · 1:1 · 21:9
Left out, Kling 4.0 follows your references, or uses 16:9 for a prompt alone. Keyframe clips always follow the first frame.
aspect_ratioRequired, up to 10,000 characters
Describe action, camera and sound, and point at inputs with @mentions. Negative prompts are not supported.
promptAlways generated
generate_audio: false returns a 400. With reference videos, keep_original_sound decides whether their original audio is kept.
keep_original_soundAsync task, polling or callback
The call returns a taskId. Poll the task or receive the result at callback_url; Kling 4.0 result files are kept for 30 days.
callback_urlTwo per-second rates; a reference video changes how the seconds are counted.
$0.471
3 seconds at 720p without video input. Reference images, subject images and keyframes are not billed separately, and failed Kling 4.0 generations are not charged.
| Kling 4.0 request | 720p | 1080p | Billed seconds |
|---|---|---|---|
| Text, reference images, image subjects, voices or keyframes | $0.157/s | $0.235/s | Output length |
| With a reference video or video subject | $0.094/s | $0.141/s | Output + max(input video, output) |
| 5 s, no video input | $0.785 | $1.175 | 5 |
| 10 s, no video input | $1.57 | $2.35 | 10 |
| 30 s, no video input | $4.71 | $7.05 | 30 |
5 s at 720p from text
$0.785
5 billed s × $0.157
10 s at 1080p
$2.35
10 billed s × $0.235
30 s single take at 720p
$4.71
30 billed s × $0.157
5 s output + 8 s reference video, 720p
$1.222
13 billed s (5 + 8) × $0.094
10 s output + 5 s reference video, 720p
$1.88
20 billed s (10 + 10) × $0.094
10 s output + 20 s reference video, 720p
$2.82
30 billed s (10 + 20) × $0.094
Why a reference video raises the total: with video input the billed seconds are the output plus the longer of the input video and the output, so a Kling 4.0 request with video always bills at least twice its length. Each clip is rounded up to whole seconds and all reference videos and video subjects are added together.
Pick Kling 4.0 for long, reference-driven takes; pick another model for the cases below.
Kling 4.0 renders up to 30 seconds in one request, so a long shot does not need stitching.
Lock a character or a product with images while a reference clip supplies motion and camera.
Pin up to 10 frames and let Kling 4.0 fill in the motion between them.
Keep a product or character consistent from your images while copying a camera move from a reference video.
Kling 4.0 stops at 1080p; Vidu Q4 on APIMODELS goes up to 4K.
Audio is always generated; with reference videos the only audio switch is keep_original_sound.
Kling 4.0 has no negative prompt field, no multi-shot storyboard and no video extension.
If length and audio matter less than price, Vidu Q4 at $0.0887/s for 720p costs less per second.
Six ways teams put Kling 4.0 to work.
Text, references and keyframes in one integration
Offer every Kling 4.0 mode from the same endpoint, with a 720p preview and a 1080p export.
Approved first frame, pack shot at the end
Start a Kling 4.0 clip from an approved key visual and end on the pack shot as the last frame.
Subjects and voices keep the cast steady
Send the cast as subjects and voices in every request so faces and voices hold across 30-second Kling 4.0 scenes.
16:9 · 9:16 · 1:1 · 21:9
Render the same Kling 4.0 concept for every feed by changing one field.
Product photos plus a hand-held reference
Turn product photos into demos, with a reference video for the hand-held camera feel.
Async tasks with callbacks
Queue Kling 4.0 jobs and collect results via callback_url; the apimodels MCP server exposes video generation and task polling to agents.
Small habits that keep Kling 4.0 jobs predictable in production.
taskId + polling or callback_url
Store the taskId and poll every 10–15 seconds, or let callback_url notify you; never hold a web request open for a Kling 4.0 render.
Billed = output + max(input, output)
Trim reference clips to the part you need: input video longer than the output adds to the Kling 4.0 bill second by second.
Files are kept for 30 days
Copy finished Kling 4.0 videos to your own storage; the link stops working once the file is deleted.
@image_3 needs a third image
Check every @mention against your arrays before calling; a Kling 4.0 request with a dangling mention returns a 400.
Frames ≤ duration + 1, max 10
A 5-second Kling 4.0 keyframe clip takes at most 6 frames; lengthen the clip before adding more.
Shows billed seconds before you run
The playground on this page computes the Kling 4.0 price, including reference-video seconds, before you click Generate.
From API key to downloaded video in four steps.
Sign up at apimodels.app and create a key in the console. One key and one USD balance cover Kling 4.0 and every other model; top-ups start at $10.
POST /v1/video/generations with model kling-4.0, a prompt and your inputs, using the key as a Bearer token.
Poll GET /v1/video/generations?task_id= until the state is completed, or pass callback_url and receive the result.
Read the file URL from resultUrls and save it within 30 days.
curl -X POST https://api.apimodels.app/v1/video/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kling-4.0",
"prompt": "@Mia walks through the market from @image_1, the camera tracking her like @video_1, and says: Fresh figs today!",
"image_urls": ["https://example.com/market.jpg"],
"video_urls": ["https://example.com/tracking-shot.mp4"],
"elements": [{
"name": "Mia",
"image_urls": ["https://example.com/mia-front.png", "https://example.com/mia-side.png"],
"tag": "character"
}],
"duration": 10,
"resolution": "1080p",
"aspect_ratio": "16:9"
}'
# { "code": 200, "msg": "success", "data": { "taskId": "clxxx", "state": "pending" } }
curl "https://api.apimodels.app/v1/video/generations?task_id=TASK_ID" -H "Authorization: Bearer YOUR_API_KEY"
# { "code": 200, "data": { "state": "completed", "resultUrls": ["https://r2.apimodels.app/videos/clxxx.mp4"] } }All prices are the per-second rates on apimodels.app today. Kling 4.0 is the one for 30-second takes, the widest reference set and keyframe sequences; Kling V3 Omni is the pick for editing a clip while keeping its sound, and Vidu Q4 for 4K or the lowest per-second price.
| Kling 4.0 | Kling V3 | Kling V3 Omni | Vidu Q4 | |
|---|---|---|---|---|
| Clip length | 3-30 s | 3-15 s | 3-15 s | 3-16 s |
| Max resolution | 1080p | Standard / pro tiers | Standard / pro tiers | 4K |
| Text-to-video | Yes | Yes | Yes | No |
| Image inputs | Up to 10 reference images, or first + up to 8 middle + last frame | First frame | Image and subject input | 1 first frame, or up to 15 reference images |
| Video input | Up to 5 reference videos plus up to 3 video subjects | No | Reference video and editing | No |
| Audio | Always on; keep_original_sound for reference videos | Optional | Optional; keep original sound when editing | On by default, can be turned off |
| Price per second | 720p $0.157 · 1080p $0.235; with video $0.094 · $0.141 per billed second | $0.12 standard; $0.18 with audio | $0.088 standard; $0.117 with audio | 540p $0.042 · 720p $0.0887 · 4K $0.3639 |
Kling 4.0 is a Video Generation API provided by Kling. Kling 4.0 is the newest video model in Kuaishou's Kling family, and every clip it makes comes with sound generated in the same pass. On APIMODELS it is one model id, kling-4.0 (kling-4-0 and kling-v4 also work), on the unified video endpoint, with two modes chosen from what you send. Reference-to-video covers plain text-to-video and adds up to 10 reference images in image_urls, mentioned in the prompt as @image_1, @image_2 and so on; up to 5 reference videos in video_urls (@video_1 and on), each 2 to 30 seconds and 30 seconds in total; up to 7 named subjects in elements, each either an image subject of 2 to 4 views with the main view first or a single-shot character video of 3 to 8 seconds, mentioned as @Name; and up to 3 named voice clips in voices, also mentioned as @Name. One request takes at most 15 materials and at most 3 video subjects, and keep_original_sound keeps the original audio of the reference videos. Keyframe-to-video takes first_frame_url (image and image_url are accepted too), up to 8 middle frames in keyframe_urls and last_frame_url, mentioned as @first_frame, @keyframe_1 and @last_frame; the total number of frames is capped by the duration, from 4 frames at 3 seconds to 10 frames at 9 seconds or longer, and the clip follows the first frame's aspect ratio. Frames cannot be combined with references, and every @mention has to point at an input in the same request; otherwise the request is rejected and nothing is charged. Both modes take a required prompt of up to 10,000 characters, duration as any whole number of seconds from 3 to 30 (default 5), resolution 720p (default) or 1080p, and aspect_ratio auto, 16:9, 9:16, 1:1 or 21:9. Sound is always generated; generate_audio: false returns a 400. Billing is per billed second. Without video input the billed seconds equal the output length, at $0.157/s for 720p and $0.235/s for 1080p, so a 5-second 720p clip is $0.785 and a 10-second 1080p clip $2.35. With reference videos or video subjects the rate is $0.094/s (720p) or $0.141/s (1080p), and the billed seconds are the output length × 2 when the input video is no longer than the output, or input + output when it is longer: a 10-second 720p clip with a 5-second reference video is 20 billed seconds, $1.88, and with a 15-second reference video 25 billed seconds, $2.35. Failed generations are not charged. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: 720p · per second: $0.157, 720p · 5s: $0.785, 720p · 10s: $1.57, 720p · 30s: $4.71, 1080p · per second: $0.235, 1080p · 5s: $1.175, 1080p · 10s: $2.35, 1080p · 30s: $7.05, Reference video · 720p · per billed second: $0.094, Reference video · 1080p · per billed second: $0.141.
Related model: Kling V3 Omni — when you edit an existing clip and want to keep its original sound — from $0.088/s, $0.117/s with audio
Related model: Vidu Q4 — for 4K or a lower per-second price — 540p from $0.042/s, 720p $0.0887/s, 3-16 seconds
Related model: FLUX 3 Max — when frames must land on exact timestamps — up to 10 keyframes placed by time or frame index, 5-20 seconds, from $0.221/s
Related model: Kling V3 — the previous Kling generation for plain text-to-video or a first frame, 3-15 seconds — from $0.12/s
Quickly generate brand promotion videos for ad campaigns and social media marketing.
Create compelling short-form video content for platforms like TikTok, Instagram, and YouTube.
Generate product feature demonstrations and tutorials to improve user conversion.
Produce course explanations, knowledge explainers, and training videos at an affordable price.
Kling 4.0 is available through APIMODELS at: 720p · per second: $0.157, 720p · 5s: $0.785, 720p · 10s: $1.57, 720p · 30s: $4.71, 1080p · per second: $0.235, 1080p · 5s: $1.175, 1080p · 10s: $2.35, 1080p · 30s: $7.05, Reference video · 720p · per billed second: $0.094, Reference video · 1080p · per billed second: $0.141. Billing is pay-as-you-go — you only pay for what you generate.
Sign up at APIMODELS, get your API key, and call our unified API endpoint. We provide detailed API documentation with code examples in cURL, Python, and Node.js.
APIMODELS offers the same Kling 4.0 model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
Kling 4.0 is Kuaishou's flagship video model, the newest generation of its Kling family. It turns a prompt, reference material or a set of keyframes into a 3 to 30 second clip at 720p or 1080p, with sound generated in the same pass. Kuaishou highlights steadier motion in fast action and in complex camera moves such as tracking and orbit shots, better lip sync, native two-channel stereo audio, speech in several languages and accents, and clearer on-screen text and logos. On apimodels.app it is one model id, kling-4.0, on POST /v1/video/generations, from $0.157 per second.
There are two rates. Without video input (text, reference images, image subjects, voices or keyframes) you pay $0.157 per second at 720p or $0.235 at 1080p for the length you request: 5 seconds at 720p is $0.785, 10 seconds at 1080p $2.35, and the smallest request, 3 seconds at 720p, $0.471. With a reference video or video subject the rate is $0.094 (720p) or $0.141 (1080p) per billed second, where billed seconds = output + max(input video, output): 10 seconds of output with a 5-second reference is 20 billed seconds, $1.88; with a 20-second reference it is 30 billed seconds, $2.82. Reference images, subject images and keyframes are not billed separately, and failed generations are not charged.
Because the billed seconds at least double. With video input the billed seconds are the output length plus the longer of the input video and the output, so a 5-second 720p clip with a reference video of 5 seconds or less bills 10 seconds at $0.094, which is $0.94, against $0.785 for the same clip without video; with an 8-second reference it bills 13 seconds, $1.222. Each clip is rounded up to whole seconds and all reference videos and video subjects are added together, so trimming reference clips to the part you need is the simplest way to keep the bill down.
Up to 30 seconds in a single take. duration takes any whole number of seconds from 3 to 30 (default 5) in both modes, so a long shot does not have to be stitched together from shorter clips. In keyframe mode the duration also caps the number of frames: 4 frames at 3 seconds, one more for each extra second, and 10 from 9 seconds up. A 30-second clip without video input costs $4.71 at 720p and $7.05 at 1080p.
resolution is 720p (default) or 1080p. aspect_ratio is auto, 16:9, 9:16, 1:1 or 21:9; left out, it follows your references, or 16:9 for a prompt alone, and in keyframe mode the clip always follows the first frame. Kuaishou has announced 4K and 10-bit HDR for Kling, but neither is available yet; Vidu Q4 on apimodels.app outputs up to 4K today.
Yes, on every clip and at no extra charge. Sound, including dialogue and lip-synced speech, is generated together with the picture; Kuaishou describes the audio as native two-channel stereo, with speech in several languages and accents. To shape a voice, add a named clip in voices or bind voice_url to a character subject, and mention it with @Name. With reference videos, keep_original_sound (default false) keeps their own audio instead.
Up to 15 materials in total: at most 10 reference images in image_urls, 5 reference videos in video_urls (2 to 30 seconds each, 30 seconds combined with video subjects), 7 subjects in elements (no more than 3 of them videos) and 3 voices in voices. Images usually set who and what appears, such as a character, a product, the composition or a style; reference videos carry motion, camera movement and pacing. The prompt points at them with @image_1, @video_1 and @Name. Numbering follows array position, so reordering image_urls changes what @image_2 refers to, and a mention with no matching input returns a 400 before anything is charged.
Send first_frame_url, up to 8 middle frames in keyframe_urls in time order and last_frame_url, and describe what happens between them using @first_frame, @keyframe_1 and @last_frame. All frames together may not exceed 10, or the duration plus one (3 seconds takes 4 frames, 9 seconds or more takes 10). The output follows the first frame's aspect ratio. Frames cannot be mixed with reference images, videos, subjects or voices in the same request, and keyframe requests are billed at the no-video rate, $0.157 per second at 720p.
Not today, for all three. Audio is always generated and generate_audio: false returns a 400; with reference videos the only audio switch is keep_original_sound. There is no negative_prompt field, no multi-shot storyboard and no video extension, and output stops at 1080p, with 4K and HDR not available yet. If the lowest price per second matters more than length or audio, a model such as Vidu Q4 at $0.0887 per second for 720p is the better fit.
Create an account at apimodels.app and generate an API key in the console; one key and one USD balance cover every model. Then POST to https://api.apimodels.app/v1/video/generations with the key as a Bearer token: curl -X POST https://api.apimodels.app/v1/video/generations -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"model": "kling-4.0", "prompt": "The chef in @image_1 flips a pancake and laughs", "image_urls": ["https://example.com/chef.png"], "duration": 5}'. The response holds a taskId: poll GET /v1/video/generations?task_id= until the state is completed, or pass callback_url, then download the file within 30 days.
Not entirely. Eligible new accounts get $0.10 of trial credit, but the smallest Kling 4.0 request, 3 seconds at 720p, costs $0.471, so a first test needs a top-up; top-ups start at $10. The playground on this page shows the price before you click Generate, including the billed seconds when you add a reference video, and failed generations are not charged.
On APIMODELS, Kling 4.0 runs alongside 60+ models on one API key and one balance, so choosing is about fit, not lock-in. It supports Text to Video, Reference to Video, Keyframe to Video, Up to 10 Images + 5 Videos, Named Subjects & Voices, Native Audio, 720p / 1080p, 3-30s, From $0.157/s, and you can weigh it on price and capability against other Video Generation models, then switch by changing a single model-name string — no new account or integration. Browse every Video Generation option with live pricing at apimodels.app/models.
Kling 4.0 supports: Text to Video, Reference to Video, Keyframe to Video, Up to 10 Images + 5 Videos, Named Subjects & Voices, Native Audio, 720p / 1080p, 3-30s, From $0.157/s. See the APIMODELS docs for full parameters and call examples.
Yes. APIMODELS exposes Kling 4.0 through a single unified API and one key — no separate provider accounts, and no need to handle each provider's regional network access yourself.
We support Stripe (Visa, Mastercard, and other international cards) and Alipay. Credits are available instantly after payment.