Skip to main content
Model logo

Kling 4.0 Flash API

  • Text to Video
  • Image to Video
  • Reference to Video
  • Video to Video

Kling 4.0 Flash is part of the Kling model familyView all Kling models

Upload images/videos above, then type @ in your prompt to reference them

Output

Your generated video will appear here

Updated

What is the Kling 4.0 Flash API?

The Kling 4.0 Flash API is live on UnificAlly as kuaishou/kling-4.0-flash-video, from $0.0756 per second of video. Kling 4.0 Flash is the fast model in Kuaishou's Kling 4.0 lineup. It generates 3 to 20 second videos at 720p with two-channel stereo audio, from a prompt, a start frame, up to 10 reference images, elements, or a reference video. It opened in the Kling app for early access on September 28, 2026, and has been live on UnificAlly since November 5, 2026.

Flash is built on Kling 4.0 and takes the same request. It trades the full model's 1080p output, 30-second length, and last-frame and keyframe control for much faster renders at a lower price per second, which makes it the Kling 4.0 model for drafts and frequent creation.

Kling 4.0 Flash API at a glance

Model IDkuaishou/kling-4.0-flash-video
EndpointPOST https://api.unifically.com/v1/tasks
Request typesText-to-video, reference images and elements, image-to-video from a start frame, video transform, video reference, multi-shot
Duration3 to 20 seconds
Resolution720p (std mode only)
Aspect ratioauto, 1:1, 9:16, 16:9, 21:9
AudioTwo-channel stereo, on by default
ReferencesUp to 10 reference images and 7 elements, 15 in total, or 14 alongside a source video
Rate$0.0756/sec
Rate with a video element$0.1008/sec
Rate with a source video$0.1008/sec

Billing is per second of output video, with no subscription. With a source video, the Kling 4.0 Flash API bills per second of the input video. The pricing page shows the live rates.

How to use the Kling 4.0 Flash API

  1. Create an API key on the API keys page. The same key and balance work for every model on UnificAlly.
  2. Send a POST request to /v1/tasks with the model ID and an input object.
  3. Read the task_id from the response, then check the task until it is finished, or pass a callback URL and wait for the result.
curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "kuaishou/kling-4.0-flash-video",
    "input": {
      "prompt": "A street food vendor flips noodles in a flaming wok at night, sizzling oil and crowd chatter",
      "duration": 5,
      "aspect_ratio": "16:9"
    }
  }'

The response returns the task ID straight away:

{
  "code": 200,
  "success": true,
  "data": {
    "task_id": "abc123def456",
    "status": "processing"
  }
}

Check the task with the same key. The finished task carries the video URL.

curl https://api.unifically.com/v1/tasks/{task_id} \
  -H "Authorization: Bearer YOUR_API_KEY"

Kling 4.0 Flash API parameters

ParameterTypeDefaultWhat it does
video_modestringnoneInput mode: elements, start_end_frame, transform, or video_reference. Omit for text-to-video
promptstringnoneText prompt. Use @Image1, @Element1 to reference inputs. Cannot be combined with multi_shots
modestringstdstd (720p) only. pro and 4k are rejected
durationinteger3Video length in seconds, 3 to 20. Ignored when video_url is set, where the source video's length is used
aspect_ratiostringnoneauto, 1:1, 9:16, 16:9, or 21:9. Must be auto when video_url is set. When omitted, the server picks auto with a source video or a start frame, otherwise 16:9
native_audiobooleantrueGenerates two-channel stereo audio with the video
keep_audiobooleantrueKeeps the original audio of the source video in transform and video_reference modes
image_urlsstring[]noneUp to 10 reference image URLs. Refer to them as @Image1, @Image2 in prompt
start_image_indexintegernone1-based index into image_urls that marks which reference image anchors the first frame
elementsobject[]nonePersistent subject or object assets, up to 7, of which at most 3 can be video elements. Refer to them as @Element1, @Element2 in prompt
start_frame_urlstringnoneFirst frame image URL for start_end_frame mode
video_urlstringnoneSource video URL for transform and video_reference modes. Billed per second of the source video at the source-video rate
multi_shotsarraynone2 to 6 shots, each with a prompt and a duration, 3 to 20 seconds in total. Cannot be combined with prompt. Not available in start_end_frame mode
nsfw_checkbooleanfalseChecks the prompt and input images for NSFW content before generating

The Kling 4.0 Flash API documentation has the full reference, with the rules for each input mode.

Key features of Kling 4.0 Flash

  • About 3.5 times faster than Kling 4.0. A 5-second 720p video with audio rendered in 64 seconds in our test, against 221 seconds on the full model.
  • 3 to 20 seconds per generation. Longer than any Kling 3.0 model, with room for a full social video in one pass, or 2 to 6 multi-shot scenes.
  • Stereo audio on by default. Dialogue, effects, and ambience come out with the video.
  • Omni Reference on the fast model. Up to 10 reference images and 7 elements, 15 references in total, so the people and places from your photos carry into the video.
  • Image-to-video and video reference. Start from a single frame, or pass a source video to restyle it or copy its motion.
  • Five aspect ratios. 16:9, 9:16, 1:1, 21:9, or auto.

What Kling 4.0 Flash is best for

  • Prompt drafts

    Test a scene at 720p in about a minute before paying for a full Kling 4.0 render.

  • Social video

    Up to 20 seconds with sound covers a full Reel or Short in one generation.

  • Character scenes from references

    Omni Reference keeps the faces and location from your reference photos.

  • High-volume pipelines

    Short render times and a lower rate per second matter most when you generate many videos a day.

Kling 4.0 Flash use cases

Run a batch of short concept drafts for a campaign, pick the strongest one, then re-render it on the full Kling 4.0 model at 1080p. Feed two portraits and a location photo into Omni Reference to produce a short scene with the same people in the same place. Animate a product still from a start frame into a vertical video for social feeds. And for apps that generate video on demand, Flash is the model to default to when the output is watched on a phone at 720p anyway and users are waiting for the result.

Kling 4.0 Flash limitations

Flash outputs 720p only, in 8-bit SDR, with no 1080p, 4K, or HDR option. It takes a start frame but no last frame or middle keyframes; those stay on the full Kling 4.0 model. The 20-second ceiling sits below the full model's 30 seconds. The speed figure comes from one side-by-side test, and render times move with Kling's queue. No public leaderboard ranks Flash yet: the Arena text-to-video and image-to-video boards and the Artificial Analysis video leaderboard have no row for it as of November 5, 2026.

Kling 4.0 Flash vs Kling 4.0

Both come from the same Kling 4.0 release and take the same request. The full model generates 3 to 30 seconds at 720p or 1080p, with a first, middle, and last frame, at $0.126/sec per second at 720p. Flash covers 3 to 20 seconds at 720p from $0.0756/sec per second, and rendered the same 5-second prompt in 64 seconds against 221. Use Flash to iterate fast and keep cost down, and move to Kling 4.0 for the final render when you need resolution, length, or keyframe control.

Kling 4.0 Flash vs Kling 3.0 Turbo

Kling 3.0 Turbo is the speed model of the previous generation. It runs 3 to 15 seconds at 720p Standard or 1080p Pro, text-to-video or image-to-video, from $0.1008 per second, without multi-reference input. Flash goes to 20 seconds, adds Omni Reference, elements, and reference video, and stops at 720p, from $0.0756 per second. Pick Flash for reference-driven drafts; Turbo keeps the edge when you need a fast 1080p render.

What Kling 4.0 Flash can do

3.5x faster than Kling 4.0 on the same prompt

We sent the same 5-second 720p prompt to Kling 4.0 Flash and Kling 4.0 at the same moment on November 5, 2026. Flash returned the finished video with audio in 64 seconds; the full model took 221 seconds.

Flames, faces, and sizzle in one fast pass

A night street-food scene from text alone: the wok flares up, the flames light the cook and the crowd behind him, and the sizzle and crowd chatter come out in stereo with the video.

Two portraits and a place become one scene

Omni Reference on the fast model: a portrait of a man, a portrait of a woman, and a photo of a clifftop become a 15-second scene of the two meeting on the cliff edge, laughing and holding hands. Both faces and the landscape match the references.

Fast martial arts from one reference still

One reference still of two karate fighters in a wooden dojo becomes a fast exchange of punches, kicks, and a leaping strike, with the gi, belts, and warm window light holding through the action.

More API examples

The same endpoint takes reference images, a start frame, or a source video. Each request below returns a task ID in the same way.

Reference images request

Set video_mode to elements and name each image in the prompt in array order.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "kuaishou/kling-4.0-flash-video",
    "input": {
      "video_mode": "elements",
      "prompt": "@Image1 and @Image2 meet on the cliff edge in @Image3, laughing",
      "image_urls": [
        "https://example.com/man.jpg",
        "https://example.com/woman.jpg",
        "https://example.com/cliff.jpg"
      ],
      "duration": 10
    }
  }'

Image-to-video request

Set video_mode to start_end_frame and pass the first frame. The frame sets the aspect ratio.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "kuaishou/kling-4.0-flash-video",
    "input": {
      "video_mode": "start_end_frame",
      "prompt": "The camera slowly pushes in as the lights come on",
      "start_frame_url": "https://example.com/street.jpg",
      "duration": 5
    }
  }'

Video reference request

Pass a source video in video_url. The output takes the source video's length.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "kuaishou/kling-4.0-flash-video",
    "input": {
      "video_mode": "video_reference",
      "prompt": "Same camera move, but in a snowy forest",
      "video_url": "https://example.com/source.mp4"
    }
  }'

FAQs

People also ask about Kling 4.0 Flash

Kling 4.0 Flash is the fast, lower-cost model in Kuaishou's Kling 4.0 lineup. It generates 3 to 20 second videos at 720p with stereo audio from a prompt, a start frame, up to 10 reference images, elements, or a reference video.

Sign up on UnificAlly and create a key on the API keys page. One key and one balance cover Kling 4.0 Flash and every other model, with no subscription.

Send a POST request to https://api.unifically.com/v1/tasks with the model ID kuaishou/kling-4.0-flash-video and an input object holding the prompt, duration, and aspect ratio. The response returns a task ID. Check /v1/tasks/{task_id} or use a callback URL to get the finished video.

Yes. Kling 4.0 Flash is live on UnificAlly as kuaishou/kling-4.0-flash-video since November 5, 2026, with a browser playground on this page and the same API key and balance as every other model.

$0.0756/sec per second of video, so a 5-second video costs five times that. With a reference video or a video element the rate is $0.1008 per second, billed on the length of the source video. There is no subscription.

In our test on November 5, 2026, Kling 4.0 Flash rendered a 5-second 720p video with audio in 64 seconds. The full Kling 4.0 model took 221 seconds on the same prompt sent at the same moment, so Flash was about 3.5 times faster.

3 to 20 seconds per generation, or 2 to 6 multi-shot scenes that add up to 3 to 20 seconds. The full Kling 4.0 model goes to 30 seconds.

720p only, in 8-bit SDR, in 16:9, 9:16, 1:1, 21:9, or auto. For 1080p output, use the full Kling 4.0 model.

It takes a start frame for image-to-video. A last frame and middle keyframes are only on the full Kling 4.0 model, so the API rejects end_frame_url and keyframe_urls on Flash.

Not yet. As of November 5, 2026, the Arena text-to-video and image-to-video boards and the Artificial Analysis video leaderboard have no Kling 4.0 Flash row, and no Kling 4.0 row either.

Use Flash for drafts, social video, and volume work where 720p is enough and turnaround matters. Use Kling 4.0 when you need 1080p, videos longer than 20 seconds, or last-frame and keyframe control.