# Kling: Kuaishou's video model — what 3.0 and 4.0 can do

> Kling is a family of video models from the Chinese company Kuaishou. Kling 3.0 launched on February 5, 2026: clips up to 15 seconds in 720p and 1080p, several shots in one generation, native audio in many languages, and characters from photos and video. On September 28, 2026, Kling announced version 4.0 with generation up to 30 seconds, still in early access. In AI2Face, Kling is available in the generator.

Developer: Kuaishou · Released: 2026-02-05 · Updated 2026-10-01
URL: https://ai2face.com/models/kling

## Key facts

| | |
|---|---|
| Developer | Kuaishou (Kling AI) |
| Versions | Kling 3.0 and 3.0 Omni (February 5, 2026), 3.0 Turbo, Kling 4.0 (announced September 28, 2026) |
| What it does | Video with sound from text, images, video and audio; clip editing |
| Duration | 3.0 — up to 15 seconds; 4.0 — up to 30 seconds (claimed) |
| Resolution | 3.0 — 720p and 1080p |
| Several shots | A storyboard with duration, shot size, angle and camera move for each shot |
| References | Up to 7 images; a character from a 3–8 second video, voice included |
| Audio | Native — speech in different languages, dialects and accents, plus music and effects |
| In AI2Face | Yes |

## What is Kling

Kling is a line of video models from the Chinese company Kuaishou, which also runs a short-video platform of the same name. As of February 2026, the company reported that more than 60 million creators had used Kling, more than 600 million clips had been generated, and more than 30,000 companies had worked with the model.

**Kling 3.0** launched on February 5, 2026. Kuaishou named the main improvements: character stability, photorealism, clips up to 15 seconds, and native audio in different languages, dialects and accents. The model is built as all-in-one: it understands text, images, audio and video, and it can both generate and edit a clip in one process. **Kling 3.0 Omni** works alongside it for references and edits. A faster **Kling 3.0 Turbo** came out later.

**Kling 4.0** was announced on September 28, 2026 — more on that below.

## What Kling can do

**Several shots in one clip.** Kling 3.0 builds a storyboard: for each shot you can set the duration, shot size, angle, action and camera move. The total length is up to 15 seconds.

[A chef chops basil and tomatoes, tosses pasta in a flaming pan, grates parmesan and sets the plate on the pass with a smile](https://ai2face.com/_astro/media/models/kling/multi-shot.mp4 "Four shots from one prompt: close-up, medium, slow close-up and wide. Kling 3.0 Pro, 15 s")

**Speech and sound.** Lines in quotes are spoken with synced lips, together with music and background.

[A street violinist in a grey coat plays under a stone arch in the evening, lowers the violin and speaks to the camera](https://ai2face.com/_astro/media/models/kling/dialogue.mp4 "A melody, a line to the camera and street noise in one generation. At the end the bow slightly detaches from the hand. Kling 3.0 Pro, 10 s")

**Physics and motion.** The model holds inertia and body weight: jumps, runs and landings.

[A parkour athlete runs across rooftops at sunset, jumps the gap between buildings and lands as the camera flies alongside](https://ai2face.com/_astro/media/models/kling/motion.mp4 "A jump across a gap with a camera move. Kling 3.0 Pro, 10 s")

**First and last frame.** Give Kling two pictures and it builds the motion between them.

[A woman with red curls stands up from a blanket at an autumn picnic, puts on a coat and a beret and brings a red maple leaf to her face](https://ai2face.com/_astro/media/models/kling/first-last-frame.mp4 "Frames from our autumn photo shoot as the start and the end. The face holds, but in the middle the clothes melt together. Kling 3.0 Pro, 9:16, 8 s")

**Character elements.** In Kling 3.0 Omni you can upload up to 7 images, build a character from 4 angles, or extract one from a video of 3–8 seconds — appearance and voice together. An audio recording of 3 seconds or longer binds a voice to the character.

**Motion Control.** A motion-transfer mode: a dance, gesture or walk from a reference video is applied to a character in a photo.

## Kling 4.0: what is known

On September 28, 2026, Kling announced 4.0. The official site describes the model as entirely new: stable, dynamic motion and immersive sound for professional-level production. The developer platform lists four features coming to the API: dynamic motion, Omni references, native 30-second generation and multiple keyframes.

Only the lighter **Kling 4.0 Flash** is available so far, in early access. The full version and the API are marked as coming soon. Press reports, citing the company, give more specific figures — 4K, HDR, a reference count — but we did not find them in Kling's official materials. We will update this page when 4.0 is fully released.

## What Kling is good for

- **Short stories and ads up to 15 seconds** with several shots and no editing.
- **Clips with speech** in different languages and accents.
- **Dances and movement from a reference** through Motion Control.
- **Action and sports** — the model holds momentum well.
- **A series with one character** built from photos or video.

## Kling and similar models

| | Kling 3.0 | [Veo 3.1](/models/veo-3-1) | [Seedance 2.5](/models/seedance) | [Gemini Omni 1.1 Flash](/models/gemini-omni) |
|---|---|---|---|---|
| Developer | Kuaishou | Google DeepMind | ByteDance Seed | Google DeepMind |
| Length per generation | up to 15 s | 4, 6, 8 s | up to 30 s | 3–10 s |
| Resolution | 720p, 1080p | up to 4K | not disclosed | up to 4K (upscale) |
| Several shots | yes, storyboard | via timecodes in the prompt | yes | yes |
| References | up to 7 photos, a character from video with voice | up to 3 images | 30 photos, 10 videos, 10 audio | images and up to 3 videos |
| Motion transfer | Motion Control | no | from a video reference | from a video reference |
| In AI2Face | yes | yes | yes | yes |

Kling 3.0 is a strong choice for 10–15 second clips with a storyboard and speech, and Motion Control is still the most convenient way to transfer movement. Veo 3.1 gives 4K and better sound in a short clip. Seedance 2.5 makes the longest clips. Gemini Omni is the most flexible with inputs and edits.

## How to write prompts for Kling

- **Number the shots:** "Shot 1: close-up of the hands… Shot 2: medium shot…". For each one, give the action and the camera.
- **Put spoken lines in quotes,** and note who is speaking and in what tone.
- **Put sound on its own line:** music, effects, background.
- **Name the physics in words:** "realistic momentum", "landing in a roll" — the model holds the motion better.
- **For a transition between two frames,** describe what happens in between: "stands up, puts on the coat, picks up the leaf".

```
Multi-shot cooking story in four shots. Shot 1: close-up of the chef's hands chopping
fresh basil and tomatoes on a wooden board. Shot 2: medium shot, the chef tosses
pasta in a hot pan and flames flare. Shot 3: slow-motion close-up as parmesan
is grated over the plate. Shot 4: wide shot of a warm trattoria kitchen, the chef
puts the plate on the pass and smiles. Sound: knife on wood, pan sizzle, kitchen
ambience, soft Italian guitar. Photorealistic, warm light
```

## How to use Kling in AI2Face

Kling is available in the AI2Face generator: clips with sound, a storyboard and motion transfer. The cost is shown before you start.

For other jobs in the same generator:

- **[Veo 3.1](/models/veo-3-1)** — clips of 4–8 seconds up to 4K with speech and effects, photo animation, first and last frame.
- **[Gemini Omni Flash](/models/gemini-omni)** — up to 10 seconds, a scene from several references, first and last frame.

Build a short 4K clip from two Veo 3.1 videos of the same character, using a reference. Both models run in the same generator, and the cost is shown before you start.

## Limitations

- **Glitches in small details.** In our tests, objects in the hands sometimes detach (the violinist's bow), and in two-frame transitions the clothes can melt from one outfit into another.
- **3.0 resolution goes up to 1080p.** 4K is not officially claimed for Kling 3.0.
- **Kling 4.0 is still in early access,** and only as Flash.
- **Access.** Officially, the Kling website, the app and the API. Paying for the subscription directly is difficult in some countries.

## Conclusions

Kling 3.0 is one of the strongest video models for short stories: a multi-shot storyboard, speech with synced lips and solid physics in one generation of up to 15 seconds. Motion Control makes it the best fit for dances and movement from a reference. Kling 4.0 promises clips twice as long — we are watching for the full release. Kling is available in AI2Face. If you need 4K or better sound in a short clip, take [Veo 3.1](/models/veo-3-1) or [Gemini Omni](/models/gemini-omni) — they are in the same generator.

## FAQ

### What is Kling?

A video model from the Chinese company Kuaishou, which also runs a short-video platform. Kling makes clips with sound from text, photos and video. According to Kuaishou in February 2026, more than 60 million creators had used it.

### What is new in Kling 3.0?

Clips up to 15 seconds instead of 10, native audio in many languages, a storyboard of several shots in one clip, more stable characters and photorealism. All of that sits in one model that understands text, images, audio and video.

### When will Kling 4.0 be released?

Kling announced 4.0 on September 28, 2026. The lighter Kling 4.0 Flash is in early access now. The full version and the API are marked as coming soon. The announcement claims generation up to 30 seconds, several keyframes and Omni references.

### What is Kling Motion Control?

A motion-transfer mode. The model takes a person's movement from a reference video and applies it to a character in a picture. That is how a photo is animated to a dance or a gesture from a real clip.

### Is Kling available in AI2Face?

Yes. Kling is available in the AI2Face generator, and the cost is shown before you generate. For 4K and short clips with sound, Veo 3.1 and Gemini Omni Flash are in the same generator.

## Sources

- https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-launches-30-model-ushering-era-where-everyone-can-be
- https://kling.ai/quickstart/klingai-video-3-omni-model-user-guide
- https://kling.ai/dev
- https://kling.ai/
- https://ai.google.dev/gemini-api/docs/veo
- https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5