Kling 4 โ
โ Back to Kling for all versions and extended capabilities.
Endpoint: POST https://vod.intl.tencentcloudapi.comAction: CreateAigcVideoTaskInternational API reference: CreateAigcVideoTask
Kling 4 ships three versions: 4.0 (flagship base model, text-to-video / image-to-video), 4.0-Omni (all-modality model that adds omni-reference video generation), and 4.0-Flash (built for frequent creation โ faster generation at a lower price per second). Compared with the 3.0 series, 4.0 / 4.0-Omni generate a single clip natively at 3โ30 seconds, with first / last frame plus middle frames (multi-keyframe), single-shot / multi-shot generation, and synchronized audio.
Availability
4.0 and 4.0-Omni are callable now; subject (SubjectInfos) features on 4.0-Omni opened on 2026-10-24. 4.0-Flash launched on 2026-11-05.
Flash is not simply "a cheaper 4.0" โ its capabilities are trimmed
4.0-Flash supports text-to-video, image-to-video and omni-reference video generation, with the same reference budget as 4.0 (10 images, 5 videos, 15 inputs per task combined). But it has two hard limits: a 20-second ceiling (4.0 allows 30) and no last frame or middle frames (image-to-video takes a single first frame only). It also renders natively at 720P only; 1080P and above are produced by upscaling (on 4.0, 1080P is native).
Swapping 4.0-Flash into a 4.0 first/last-frame or middle-frame example will fail โ see Choosing a version.
Basic information โ
| Item | Value |
|---|---|
| ModelName | Kling |
| ModelVersion | 4.0 (text / image) / 4.0-Omni (omni-reference video) / 4.0-Flash (frequent creation: text, image, omni-reference) |
| Resolution | 4.0 / 4.0-Omni: 720P / 1080P / 2K (upscaled) / 4K (upscaled)4.0-Flash: 720P / 1080P (upscaled) / 2K (upscaled) / 4K (upscaled) โ native up to 720P |
| Aspect ratio | 21:9 / 16:9 / 9:16 / 1:1 |
| Duration | 4.0 / 4.0-Omni: 3โ30 seconds (default 10 s)4.0-Flash: 3โ20 seconds |
| Audio output | Supported (audio and video generated together) |
| Pricing | See the AIGC pricing guide (customer edition) |
Choosing a version โ
| Your scenario | Pick | Why |
|---|---|---|
| Batch-testing prompts, drafting, checking framing and pacing | 4.0-Flash | Fastest and cheapest; native 720P is enough to judge camera work |
Delivering a finished clip that needs native 1080P or HDR | 4.0 | On Flash, 1080P+ is upscaled; only 4.0 renders it natively, and HDR is 4.0-only |
| Precise control over where frames land (last frame, middle keyframes) | 4.0 | Flash takes a single first frame only |
| Consistency from reference video / images / subjects / voices | 4.0-Omni | Subjects and voices are Omni-only |
Reference assets without subjects or voices, and native 720P is acceptable | 4.0-Flash | Image / video reference budget matches 4.0 |
A common pattern is two-stage: run several variants of the same shot on 4.0-Flash to settle on the prompt, then re-render on 4.0 / 4.0-Omni for delivery.
Version comparison โ
| Version | Capabilities | Highlights | Max reference images | Audio | Duration |
|---|---|---|---|---|---|
4.0-Omni | Text, image, first / last frame, middle frames, omni-reference (video / image / subject / voice) | All-modality model; up to 10 reference images and 5 reference videos | 10 | Original sound / voice-over / SFX | 3โ30 s |
4.0 | Text, image, first / last frame, middle frames | Flagship base model, stable quality, first choice for everyday commercial use | 10 | Supported | 3โ30 s |
4.0-Flash | Text, image (first frame only), omni-reference | Built for frequent creation: faster and cheaper; native 720P, no last frame or middle frames | 10 | Supported | 3โ20 s |
3.0-Omni (for comparison) | Text, image, reference | Previous-generation omni model | 8 | Original sound / voice-over / SFX | 3โ15 s |
Differences from Kling 3.0 โ
| Item | Kling 4.0 | Kling 3.0 |
|---|---|---|
| Positioning | 30 s native output, large motion, omni-reference, precise storytelling | Audio-video sync, stronger subject consistency, multi-shot storytelling |
| Duration | 3โ30 s | 3โ15 s |
| Dynamic range | SDR, 10-bit HDR | SDR |
| Aspect ratio | 21:9 / 16:9 / 1:1 / 9:16 | 16:9 / 1:1 / 9:16 |
| Multi-keyframe (middle frames) | Supported | Not supported (3.0 / 3.0-turbo / 3.0-Omni) |
Differences between Flash and 4.0 โ
| Item | 4.0-Flash | 4.0 |
|---|---|---|
| Positioning | Frequent creation: faster, cheaper per second | Final delivery: quality and control first |
| Duration | 3โ20 s | 3โ30 s |
| Resolution | Native 720P; 1080P / 2K / 4K via upscaling | 720P / 1080P native; 2K / 4K upscaled |
| Dynamic range | 8-bit SDR | SDR, 10-bit HDR |
| Image-to-video | Single first frame only | First / last frame + middle frames (multi-keyframe) |
| Omni-reference | Supported (10 images, 5 videos, 15 inputs total) | Supported |
Capabilities and limits โ
| Mode | Capability | 4.0 | 4.0-Omni | 4.0-Flash | Limits |
|---|---|---|---|---|---|
| Text-to-video | Single / multi-shot, audio and video together | โ | โ | โ | โ |
| Image-to-video | First frame | โ | โ | โ | โ |
| Image-to-video | Last frame (first / last frame) | โ | โ | โ | Flash takes a single first frame only |
| Image-to-video | Multi-keyframe (middle frames) | โ | โ | โ | Not supported on Flash |
| Image-to-video | Voice reference | โ | โ | โ | Voices require 4.0-Omni omni-reference |
| Omni-reference | Video reference | โ | โ | โ | Up to 5 videos, total length โค 30 s |
| Omni-reference | Image reference | โ | โ | โ | Up to 10 images |
| Omni-reference | Subject management | โ | โ | โ | Up to 7 subjects, at most 3 of them from reference videos |
| Omni-reference | Voice reference | โ | โ | โ | Bind a voice to a subject, or use an existing voice directly |
| Omni-reference | Total references | โ | โ | โ | Images, videos, subjects and voices combined โค 15 |
| Output | Native 1080P | โ | โ | โ | Flash renders natively up to 720P (8-bit SDR); 1080P+ is upscaled |
The Flash reference budget comes from the official Kling API changelog (2026-11-04). Section 1.3 of the official Kling 4.0 User Guide lists only six items for Flash (text / image / omni-reference, 3โ20 s, 720P, 8-bit SDR) and does not state the budget separately. Confirm against your own integration testing.
Request parameters โ
| Parameter | Required | Type | Description |
|---|---|---|---|
SubAppId | Required | Integer | VOD application ID |
ModelName | Required | String | Fixed value Kling |
ModelVersion | Required | String | 4.0 / 4.0-Omni / 4.0-Flash |
Prompt | Required | String | Generation prompt. Reference assets with @image_N / @video_N / @element_N / @voice_N |
FileInfos.N | Optional | Array | Reference frames, images, videos and other resource files; fields below |
SubjectInfos.N | Optional | Array | Subject input (4.0-Omni only). Id required (fixed subject ID), Name optional |
OutputConfig | Optional | Object | Output settings: resolution, aspect ratio, duration, audio, storage mode, etc. On 4.0-Flash, Duration must not exceed 20; Resolution accepts 720P / 1080P / 2K / 4K, but only 720P is native |
ExtInfo | Optional | String | Extended parameters (escaped JSON string), e.g. the voice list voice_list (4.0-Omni only) |
FileInfos fields โ
| Field | Value | Description |
|---|---|---|
Category | Image / Video | File category; use together with Usage |
Usage | FirstFrame | First frame |
Usage | LastFrame | Last frame for first / last frame generation (not supported on 4.0-Flash) |
Usage | Reference | Reference frame / reference asset |
ReferenceType (Category=Image) | middle_image | Middle-frame image; Usage must be Reference (not supported on 4.0-Flash) |
ReferenceType (Category=Image) | refer_image | Element reference image; Usage must be Reference |
ReferenceType (Category=Video) | feature / feature_video | Feature reference video |
ReferenceType (Category=Video) | base / base_video | Video to be edited |
The middle-frame identifier is middle_image (not middle_frame). Use it together with Usage: Reference, as in the example below.
Request examples โ
Text-to-video (4.0) โ
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0",
"Prompt": "generate a car",
"OutputConfig": {
"Resolution": "1080P",
"Duration": 10,
"AspectRatio": "16:9",
"StorageMode": "Temporary"
}
}Image-to-video: first + middle + last frame (4.0) โ
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/a.png",
"Usage": "FirstFrame"
},
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/c.png",
"ReferenceType": "middle_image",
"Usage": "Reference"
},
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/b.png",
"Usage": "LastFrame"
}
],
"Prompt": "generate a car"
}Omni-reference: reference image + feature video (4.0-Omni) โ
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Omni",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/a.jpeg",
"ReferenceType": "refer_image",
"Usage": "Reference"
},
{
"Type": "Url",
"Category": "Video",
"Url": "https://<cdn>/a.mp4",
"ReferenceType": "feature_video",
"Usage": "Reference"
}
],
"Prompt": "Use the outfit from @image_1 and the motion rhythm of @video_1, photorealistic."
}Video editing + subject (4.0-Omni) โ
Subject features are available only after October 24.
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Omni",
"FileInfos": [
{
"Type": "Url",
"Category": "Video",
"Url": "https://<cdn>/a.mp4",
"ReferenceType": "base_video",
"Usage": "Reference"
}
],
"SubjectInfos": [
{
"Id": "324698887867107"
}
],
"Prompt": "Continue the action of @video_1, and @element_1 walks into the frame"
}Adding a voice (4.0-Omni) โ
Specify voices via voice_list in ExtInfo, and reference them in the prompt as @voice_N.
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Omni",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/a.png",
"ReferenceType": "refer_image",
"Usage": "Reference"
}
],
"Prompt": "A person says hello using voice @voice_1",
"ExtInfo": "{\"AdditionalParameters\":\"{\\\"voice_list\\\": [{\\\"voice_id\\\": 942193199232594917}]}\"}"
}The
voice_idmust be created first: see VOD โ Creating a custom voice (CreateAigcCustomVoice).
Text-to-video (4.0-Flash) โ
Duration must not exceed 20. Resolution accepts 1080P and above, but Flash renders natively only up to 720P โ higher resolutions are upscaled, so use 4.0 when final-delivery image quality matters.
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Flash",
"Prompt": "generate a car",
"OutputConfig": {
"Resolution": "720P",
"Duration": 10,
"AspectRatio": "16:9",
"StorageMode": "Temporary"
}
}Image-to-video: single first frame (4.0-Flash) โ
Flash accepts only a single first frame. Do not pass Usage: LastFrame or ReferenceType: middle_image โ those are available on 4.0 / 4.0-Omni only.
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Flash",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/first.png",
"Usage": "FirstFrame"
}
],
"Prompt": "the car drives away slowly, cinematic",
"OutputConfig": {
"Resolution": "720P",
"Duration": 10,
"StorageMode": "Temporary"
}
}Omni reference: reference image + feature video (4.0-Flash) โ
Reference-asset quotas match 4.0 (10 images, 5 videos, 15 items per task), but subjects (SubjectInfos) and voices (voice_list) remain 4.0-Omni only.
{
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0-Flash",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/a.jpeg",
"ReferenceType": "refer_image",
"Usage": "Reference"
},
{
"Type": "Url",
"Category": "Video",
"Url": "https://<cdn>/a.mp4",
"ReferenceType": "feature_video",
"Usage": "Reference"
}
],
"Prompt": "Take the outfit from @image_1 and the motion rhythm of @video_1, photorealistic.",
"OutputConfig": {
"Resolution": "720P",
"Duration": 10,
"StorageMode": "Temporary"
}
}Code sample (Python) โ
Replace the model version and parameters as needed. Credentials are read from the TENCENTCLOUD_SECRET_ID / TENCENTCLOUD_SECRET_KEY environment variables to avoid hard-coding.
import os
import json
from tencentcloud.common import credential
from tencentcloud.common.profile.client_profile import ClientProfile
from tencentcloud.common.profile.http_profile import HttpProfile
from tencentcloud.common.exception.tencent_cloud_sdk_exception import TencentCloudSDKException
from tencentcloud.vod.v20180717 import vod_client, models
try:
cred = credential.Credential(os.getenv("TENCENTCLOUD_SECRET_ID"), os.getenv("TENCENTCLOUD_SECRET_KEY"))
http_profile = HttpProfile()
http_profile.endpoint = "vod.intl.tencentcloudapi.com"
client_profile = ClientProfile()
client_profile.httpProfile = http_profile
client = vod_client.VodClient(cred, "", client_profile)
req = models.CreateAigcVideoTaskRequest()
params = {
"SubAppId": 123456789,
"ModelName": "Kling",
"ModelVersion": "4.0",
"FileInfos": [
{
"Type": "Url",
"Category": "Image",
"Url": "https://<cdn>/first.png",
"Usage": "FirstFrame"
}
],
"Prompt": "generate a car",
"OutputConfig": {
"StorageMode": "Permanent",
"MediaName": "kling-4-video"
}
}
req.from_json_string(json.dumps(params))
resp = client.CreateAigcVideoTask(req)
print(resp.to_json_string())
except TencentCloudSDKException as err:
print(err)Response and result query โ
The request returns a TaskId:
{
"Response": {
"RequestId": "<request-id>",
"TaskId": "<task-id>"
}
}Poll the task with DescribeTaskDetail. AigcVideoTask.Status=FINISH with ErrCode=0 means success; the video URL is in AigcVideoTask.Output.FileInfos.FileUrl.
Prompt reference syntax
Kling 4 examples reference assets as @image_1 / @video_1 / @element_1 / @voice_1; the Kling 3.0 examples on the main page use the <<<image_1>>> form.
Billing when an input video is involved
The Kling 4 series is billed on total duration, and input videos (reference videos, videos being edited, video subjects) count toward it:
- Input video longer than the output: billed as "input seconds + output seconds".
- Input video the same length or shorter: billed as "output seconds ร 2".
So when you use a reference video, the actual cost is not output duration ร unit price. Refer to the AIGC pricing guide (customer) for exact unit prices.