# Kling 4: 30-second clips, 10 keyframes and 15 references — what Kling 4.0 can do

> Kling 4 is the fourth generation of Kuaishou's video models. Early access to the lighter Kling 4.0 Flash opened on September 28, 2026, and the full Kling 4.0 launched on October 15. It makes 3–30 second clips in one generation, takes up to 10 keyframes and up to 15 references — images, videos, characters and voices — and edits up to 5 clips at once. Kling 4.0 is available in the AI2Face generator.

Developer: Kuaishou (Kling AI) · Released: 2026-10-15 · Updated 2026-10-28
URL: https://ai2face.com/models/kling-4

## Key facts

| | |
|---|---|
| Developer | Kuaishou (Kling AI) |
| Release date | Early access (Flash) — September 28, 2026; full version — October 15, 2026 |
| Duration | 3–30 seconds per generation |
| Keyframes | Up to 10 on the timeline |
| References | Up to 15 files — 10 images, 5 videos (up to 30 s), 7 characters, voices |
| Editing | Up to 5 input clips, 30 seconds in total |
| Audio | Stereo, improved lip-sync |
| Prompt | Up to 8,000 tokens |
| Coming soon | 4K, 21:9, 10-bit HDR, extension up to 2 minutes, API for businesses |
| In AI2Face | Yes, in the generator |

## What is Kling 4

Kling 4.0 is the fourth generation of video models from the Chinese company Kuaishou. An overview of the whole family is on the [Kling](/models/kling) page.

The launch came in two stages:

1. **September 28, 2026:** Kling announced 4.0 and opened early access to the lighter **Kling 4.0 Flash** for a limited group of subscribers.
2. **October 15, 2026:** the full Kling 4.0 launched, and Kuaishou issued a press release the next day.

Kling describes 4.0 as a model for professional-grade production: stable dynamic motion, immersive sound and precise control through frames and references. According to Kling, the platform has more than 60 million users and more than 600 million videos created. Kling 4.0 isn't on the Arena leaderboard yet.

## What Kling 4.0 can do

**Up to 30 seconds per generation.** Length is 3 to 30 seconds — twice the 15 seconds of [Kling 3.0](/models/kling-3-0). A long take doesn't have to be stitched from pieces.

**10 keyframes.** You can place up to 10 images on the timeline to set a character's state, scene changes and story beats. The model builds the motion between them.

![Six frames from the story of a paper boat: a child sets it in a gutter stream, it races along the gutter, drops into a canal, floats under a bridge with ducks, heads out into a river at sunset and onto the sea under the stars](https://ai2face.com/_astro/keyframes.CCMh3o1Y_1SGEyd.jpg "Six keyframes of one 30-second story — Kling 4.0 takes up to ten. Made with Nano Banana Pro")

**Omni references — up to 15 files per task:**

- up to 10 images: character photos, turnaround sheets, grids, storyboards, wireframes;
- up to 5 videos, 30 seconds in total: performance, motion, camera, pacing; white-model 3D blockouts and depth maps work too;
- up to 7 character elements, up to 3 of them from video;
- voices — so a character sounds the same every time.

![A reference board: a turnaround sheet of a courier girl in a red jacket, a photo of a red bag, a rainy neon street at night and a pencil storyboard of the courier running](https://ai2face.com/_astro/refs.DqTUwu5X_1Ttq71.jpg "How to assemble references for Kling 4.0: character, product, location and storyboard — each file has its own job. Made with Nano Banana Pro")

**Stable motion.** Kling highlights fast fights, sports and large-scale action, where models used to lose body shape. The camera can push in, pull out, pan, truck, track and orbit.

**Stereo sound.** Two-channel audio, tighter lip-sync and spatial ambience.

**Text in the frame.** Lettering, logos and emoji stay sharp through camera moves and transitions. Kling lists Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi and more.

**Editing finished video.** Up to 5 input clips (one primary), 30 seconds in total. You can add, replace or remove an object and change the background, facial expressions, colors, materials and camera angle while keeping the rest. Creative Recreation takes the pacing, composition and camera movement from a sample clip and swaps in your product or character.

**Prompts up to 8,000 tokens** — the model itself refines and breaks down long instructions.

## What works now and what's coming

| Feature | Status as of Oct 28, 2026 |
|---|---|
| 3–30 s per generation | works |
| 10 keyframes | works |
| 15 references, voices | works |
| Editing up to 5 clips | works |
| Stereo sound | works |
| 4K and 1080p, 21:9 | marked "coming soon" on the model page |
| 10-bit HDR | coming soon |
| Extension up to 2 minutes | coming soon |
| API for businesses and developers | coming soon; MCP and CLI access already available |

![An ultra-wide frame: a lone rider crosses a salt flat at sunset, the pink-clouded sky reflected in a thin layer of water](https://ai2face.com/_astro/ultrawide.DrQhWBKq_Z1dA5Lr.jpg "A 21:9 frame — a format Kling 4.0 promises to support. Made with Nano Banana Pro")

## Kling 4.0 vs Kling 3.0

| | [Kling 3.0](/models/kling-3-0) | Kling 4.0 |
|---|---|---|
| Released | February 2026 | Oct 15, 2026 |
| Length | up to 15 s | 3–30 s |
| Story control | shot-by-shot storyboard | up to 10 keyframes |
| References | up to 7 images or elements | up to 15: 10 images, 5 videos, 7 characters, voices |
| Video input | 1 clip, 3–10 s | up to 5 clips, 30 s total |
| Audio | yes | stereo |
| 4K | yes, since April 2026 | coming soon |
| Prompt | — | up to 8,000 tokens |

## What Kling 4.0 is good for

Kling suggests starting from the purpose of the video — an ad, a narrative, UGC, a product demo, a brand film.

- **30-second ads** in one take with the product from references.
- **Short stories and series** — 10 keyframes keep the plot on track, voices keep the characters consistent.
- **Action and sports** — fights, chases, competitions.
- **Recreating trends** — Creative Recreation from a sample clip with your product.
- **Editing footage** — swapping the background, an object or the angle.

## Kling 4.0 vs alternatives

| | Kling 4.0 | [Seedance 2.5](/models/seedance-2-5) | [Wan 3.0](/models/wan-3-0) | [MiniMax H3](/models/minimax-h3) |
|---|---|---|---|---|
| Developer | Kuaishou | ByteDance Seed | Alibaba | MiniMax |
| Length | 3–30 s | 4–30 s | 2–30 s | 4–15 s |
| Keyframes | up to 10 | shots or timestamps | first and last | first and last |
| References | up to 15 | up to 50 files | 10 photos, 5 videos, 5 audio, a document | up to 12 files |
| Editing | up to 5 clips | yes | yes | yes |
| Resolution | 4K coming soon | up to 1080p | up to 1080p | up to 2K |
| In AI2Face | yes | yes | yes | yes |

Three models make 30 seconds — Kling 4.0, Seedance 2.5 and Wan 3.0. Kling offers the most precise story control through keyframes, [Seedance](/models/seedance) takes the most references, and [Wan](/models/wan) builds clips from documents and has the clearest pricing.

## How to write prompts for Kling 4.0

Kling's formula is **subject + action + setting + camera + lighting + pacing + sound**.

1. **Break the 30 seconds into story beats** and place keyframes on them.
2. **Give each reference a job:** character, product, scene, style, or a video for motion and camera.
3. **Write like a director:** what's in the frame, how the camera moves, when the pacing changes.
4. **In dialogue, describe the character's expression, gesture and tone.**
5. **Check continuity across the whole clip** and fix individual elements instead of regenerating.

```
30 seconds, one story. The courier from the turnaround sheet in a red jacket, with the bag
from image 2, runs down the rainy neon street from image 3, as in the storyboard. 0–8 s:
medium shot, the camera tracks alongside. 8–16 s: a jump over a puddle, slow motion.
16–24 s: push in on the bag, raindrops on the fabric. 24–30 s: the courier stops at a door
and smiles at the camera. Sound: rain, footsteps in water, a distant synth, stereo.
```

## How to use Kling 4.0 in AI2Face

Kling 4.0 is in the AI2Face generator alongside [Kling 3.0](/models/kling-3-0), [Kling 2.6](/models/kling-2-6) and [Kling O1](/models/kling-o1).

1. Sign up — new accounts get starter credits. What Kling itself offers for free is covered in the article on [free Kling](/blog/kling-besplatno).
2. Open the generator, switch to video and pick Kling 4.0.
3. Upload keyframes and references, write a prompt and set the length.
4. The credit cost is shown before you start. Credits don't expire.

References — a character turnaround, product photos, locations — are easy to make in the same place with [Nano Banana Pro](/models/nano-banana-pro).

## Limitations

- **4K, 21:9 and HDR are still "coming soon."** For 4K right now, use Kling 3.0.
- **Extension up to 2 minutes is "coming soon."** For now the ceiling is 30 seconds.
- **The business API is "coming soon."** Officially the model runs in the Kling app and via MCP and CLI.
- **Limits within the 15 files.** Each type has its own cap: 10 images, 5 videos, 7 characters.
- **Video input — 30 seconds in total.**
- **Not ranked yet.** There are no independent quality comparisons for 4.0 so far.

## Bottom line

Kling 4.0 is the most controllable of the 30-second video models: 10 keyframes, 15 references with voices and editing up to 5 clips in one model. Use it for ads and stories where the plot has to follow your plan. While 4K and HDR are only promised, render 4K finals in [Kling 3.0](/models/kling-3-0), and if you need dozens of references, compare it with [Seedance 2.5](/models/seedance-2-5). Both, along with [Wan 3.0](/models/wan-3-0), are in the AI2Face generator.

## FAQ

### When was Kling 4.0 released?

On September 28, 2026, Kling opened early access to the lighter Kling 4.0 Flash for some subscribers, and on October 15, 2026, it launched the full Kling 4.0. Kuaishou's press release followed on October 16.

### What's new in Kling 4.0?

Clips up to 30 seconds instead of 15, up to 10 keyframes, up to 15 references per task — images, videos, characters and voices — editing up to 5 clips at once, stereo sound, prompts up to 8,000 tokens and stable motion in fast scenes.

### How long are Kling 4.0 videos?

From 3 to 30 seconds per generation. Kling also promises extending a clip forward and backward up to 2 minutes, but the model page marks that feature as coming soon.

### Does Kling 4.0 support 4K?

Kling announced 4K, the 21:9 widescreen format and 10-bit HDR, but the model page still marks them as coming soon. If you need 4K now, Kling 3.0 has had a 4K mode since April 2026.

### How is Kling 4.0 different from Kling 3.0?

Twice the length (30 seconds versus 15), 10 keyframes instead of a shot-by-shot storyboard, twice as many references plus video inputs, and stereo sound. Kling 3.0 still has the edge with its ready 4K mode.

## Sources

- https://kling.ai/blog/kling-4-ai-video-model
- https://kling.ai/feature/kling-4-ai-video-generation
- https://finance.yahoo.com/technology/ai/articles/kling-ai-officially-launches-next-185000472.html
- https://kling.ai/dev