Kling 4.0
来自 Kuaishou
Thirty seconds in a single pass, with ten keyframes to steer it.
Omni Reference keeps the cast and the room consistent.
核心能力
Thirty seconds, one pass
Kling 4.0 generates up to thirty seconds natively, so a long camera move or a whole beat of a scene no longer has to be stitched from short takes.
Up to ten keyframes
Pin character states and scene changes at the moments that matter, and the model fills the motion between them.
Omni Reference
Combine up to fifteen references — images, video clips, subjects and voices — in one request, so one cast and one set carry through.
Sound with the picture
Dialogue, ambience and sound effects are generated alongside the footage rather than added afterwards.
Prompts built for shot lists
Prompts run to eight thousand tokens, enough to write a timed sequence of beats instead of a single sentence.
720p and 1080p
Output is available at 720p and 1080p with a 21:9 option; Kuaishou lists 4K and HDR as rolling out.
技术规格
Kling 4.0
Kuaishou's video model, launched October 2026
Up to 30s
Native single-pass generation
10 keyframes
Character states and scene changes you set
15 references
Images, clips, subjects and voices via Omni Reference
720p / 1080p
4K and HDR announced as coming
21:9 to 9:16
Cinema wides through verticals
Native audio
Dialogue, ambience and effects with the picture
Flash tier
A lighter 4.0 Flash variant for quick passes
提示词示例
Timed beats
Two seconds: a woman sets a brass key on the table. Four seconds: he slides his hand across it. Six seconds: the rain on the window stops
编辑提示词Night lookout
A man in a dark coat scans a rain-soaked city street through a brass telescope, headlights sweeping the wet asphalt below
编辑提示词Long take
One continuous shot follows a waiter from the loud kitchen, through the swing door, into the quiet dining room
编辑提示词Product at dawn
Slatted morning light crosses an amber perfume bottle on a travertine block as a petal drifts down beside it
编辑提示词Roadside glow
A silver trailer diner glows under a violet dusk sky as a red convertible pulls up to the pump
编辑提示词Wingsuit run
Two wingsuit flyers skim a red desert ridge, the camera flyer calling a count over the wind
编辑提示词From a keyframe
The drummer drives a slow, heavy post-rock build, sticks crashing down on the ride cymbal as the fluorescent tube flickers; the camera pushes in slowly down the green concrete corridor
编辑提示词使用场景
Beat-timed sequences
Write the clip as a timed list of beats and the take follows it, which suits a verse that has to hit moments on the second.
One continuous take
A moving camera that stays in one shot for the whole phrase, instead of three takes cut to look like one.
Keyframed transformations
Set the first and last state and let the model animate the change, from a flipbook drawing to a full reveal.
Talking to camera
Spoken lines with accent and expression come out with the shot, which helps spoken-word intros and character cameos.
Product and still life
Slow, controlled light on a single object — packaging, a record sleeve, a bottle — with motion that stays on brief.
Action with sound effects
Explosions, engines and debris arrive with their own sound, so a spectacle shot is usable straight out of the render.
概览
Kling 4.0 is Kuaishou's fourth-generation video model, launched in October 2026. Its headline change is length: a single generation can run up to thirty seconds, with up to ten keyframes marking where the story turns and an Omni Reference input that takes up to fifteen images, clips, subjects and voices at once. Sound is produced with the picture, and prompts can be long enough to hold a whole shot list.
What the keyframes and references change
Earlier Kling versions gave you a start frame and a prompt. Version 4.0 lets you place several anchors across the take, so a character can be in one state at the first beat and another at the last, with the model resolving the travel between them.
Omni Reference is the continuity half of that. The same references feed the whole thirty seconds, which is what keeps a face, a costume or a room from drifting by the end of a long take.
Resolution and availability
Kling 4.0 outputs 720p and 1080p, with a 21:9 option. Kuaishou has announced 4K and 10-bit HDR as coming, and wider API access is rolling out, so check Kling's own pages for the current state before planning a delivery around either. A lighter 4.0 Flash tier was opened to early subscribers ahead of the full launch.
Kling 4.0 and Dimly
Kling 4.0 is in Dimly's model picker at 720p and 1080p, for clips from three to thirty seconds. In a music video each shot starts on its keyframe, with an optional last frame, and your song is laid back over the finished film. Reference images and clips work in the studio chat; reference audio does not, because Kling takes its references as pictures, clips and named voices rather than a song.
If a shot has to be lip-synced to the track, MiniMax H3 and Seedance 2.5 take a slice of the song as a reference and Kling 4.0 does not. If it is a long take that has to land on an exact frame, Kling 4.0 and Seedance 2.5 both fit.
简单的价格
先免费规划你的视频，再按发布节奏选一个计划。
Weekly Pass
专注一周：做一两支副歌短片，模型任选。
- 150 积分，有效期 7 天
- 约 2 支 15 秒副歌短片
- 全部模型与分辨率 · 同时渲染 2 个
Starter
稳定的一个月：多支副歌短片，或一首较短的歌。
- 600 积分，有效期 30 天
- 约 8 支 15 秒副歌短片
- 或一首最长 2:20 的歌
- 全部模型与分辨率 · 同时渲染 3 个
Pro
每月一首完整歌曲，还有余量重做；发行需要时可加大积分。
每月积分
- 1,000 积分，有效期 30 天
- 约 14 支 15 秒副歌短片
- 或 1 首 3 分钟完整歌曲
- 全部模型与分辨率 · 同时渲染 4 个
Creator
每月多首完整歌曲，适合稳定的发行节奏。
- 5,000 积分，有效期 30 天
- 约 72 支 15 秒副歌短片
- 或约 6 首 3 分钟完整歌曲
- 全部模型与分辨率 · 同时渲染 8 个
完整歌曲、副歌短片、歌词视频或可视化 · 所有套餐导出均无水印 · 导出权益取决于你的音频权限与供应商条款
积分包
一次付清，无需订阅。适合只做一个 AI 特效或一首歌；如果会经常做，上面的套餐更划算。
单次
一条 AI 特效视频，比如 Hotel Lobby 或 AI 双人说唱。
- 250 积分，1 年内有效
- 1 条 AI 特效视频（Hotel Lobby、AI 双人说唱）
- 或约 3 段副歌短片（15 秒）
- 导出带 Dimly 水印
- 无需订阅，不会自动续费
一首歌
一支完整的 3 分钟音乐视频，或三条 AI 特效。
- 800 积分，1 年内有效
- 3 条 AI 特效视频（Hotel Lobby、AI 双人说唱）
- 或 1 首 3 分钟完整歌曲
- 导出无水印
- 无需订阅，不会自动续费
工作室
一支 2K 的 3 分钟音乐视频并留有重拍余量，或两首歌。
- 1,600 积分，1 年内有效
- 6 条 AI 特效视频（Hotel Lobby、AI 双人说唱）
- 或约 2 首 3 分钟完整歌曲
- 导出无水印
- 无需订阅，不会自动续费
估算按 MiniMax H3 768P 计算：每个镜头 5 秒（15 秒副歌短片按 3 个镜头算）、各配一张关键帧、两张参考表、不含重做；更高分辨率和其他模型每支视频花费更多。生成前你会先看到准确报价，生成失败的积分自动退回。
常见问题
Kuaishou's fourth-generation video generation model, launched in October 2026. It generates up to thirty seconds in a single pass, supports up to ten keyframes and an Omni Reference input of up to fifteen images, clips, subjects and voices, and produces sound with the picture.
Yes. Pick Kling 4.0 in the studio's model picker, at 720p or 1080p for three to thirty seconds. A 5-second 720p clip costs 50 credits, and the studio quotes every render before it runs. It does not take a song as a reference, so lip-synced shots stay on MiniMax H3 or Seedance 2.5.
Up to thirty seconds natively in one generation, compared with the short takes of earlier Kling versions. Video-to-video editing runs three to thirty seconds.
Up to ten moments in the take where you fix the character state or scene — the model animates the motion between them. It is the long-clip counterpart to first-and-last-frame control.
720p and 1080p, with a 21:9 option. Kuaishou lists 4K and 10-bit HDR as coming soon, so confirm the current state with Kling before relying on them.
The drummer in the green corridor is a Dimly render: Kling 4.0 at 720p, three seconds, animated from one of our own keyframes with its sound as generated. The rest are excerpts of Kling 4.0 renders published by fal and Morphic for the launch, shown as examples of the model rather than as Dimly renders.