Kling 4.0 已在 Runway Dev 和 fal 两个平台上线，开发者可以通过 API 与 Kling 全系列模型一起调用。这次更新主要解决了视频生成时长、音频同步、多参考输入和运动控制的实际需求。
生成能力与输入控制
Kling 4.0 可生成最长 30 秒带原生音频的视频。原生音频指视频生成时同步产生的音轨，不是后期配音。
输入控制方面，模型支持最多 15 个参考图，这些参考可以是图像、视频或元素。同时支持最多 10 个关键帧，开发者可以通过关键帧指定视频中特定时刻的画面状态，实现对生成内容的精准控制。
技术改进
官方提到改进了运动、角色一致性和电影化叙事。运动改进指镜头移动和物体动作的流畅度，角色一致性指同一角色在视频不同帧之间保持外观特征，电影化叙事涉及镜头语言和画面节奏。具体改进程度和对比基准未公开。
可用性与定价
Kling 4.0 的 API 在 fal 平台面向所有客户开放，Runway Dev 也已上线。11 月 30 日前 Kling 4 享 20% 折扣，折扣是否适用于所有平台未说明。
适用人群是需要将视频生成集成到产品或工作流的开发者，特别是需要批量生成或自动化处理的场景。如果只需要偶尔生成视频，网页界面可能更直接。
关键帧和多参考输入的具体使用方式、原生音频的质量表现、不同参考类型的组合规则，目前没有更多公开信息。
原文地址：tools.wflynn.cn/aitools/tre…
关注公众号「YourTools」，每天推送 GitHub 趋势与开源工具速报：
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Kling 4.0 is on line on two platforms, Runway Dev and fal, and the developers can call with Kling's whole series of models through API. The update addressed the actual needs for video generation time, audio synchronization, multi-reference input and movement control.
Generating capacity and input control
Kling 4.0 can generate up to 30 seconds of video with native audio. post-productionvoiceover.
For input control, the model supports up to 15 reference maps, which can be images, videos or elements. A maximum of 10 key frames are also supported, and the developer can specify a state of image at a given moment in the video via the cleared key frame to achieve precision control over the creation of content.
Technical improvements
Official mention was made of improvements in sports, role coherence and film narratives. Motion improvement refers to the fluidity of shot movements and object movements, and role consistency refers to the maintenance of visual features between different frames of the video by the same character, with film characterizations involving shot language and graphic rhythm. Specific improvements and comparative benchmarks are not available.
Availability and pricing
Kling 4.0 API is open to all clients on the fal platform and Runway Dev is on line. Kling 4 had a 20 per cent discount until 30 November, and it was not specified whether it applied to all platforms.
The applicable population group is the developer who needs to integrate video generation into the product or workflow, in particular the scenes that require bulk generation or automated processing. The web interface may be more direct if only occasional videos are generated.
Specific usage of key frames and multiple reference inputs, quality performance of primary audio, combination rules for different types of reference are not available for further public information.
Original address:tools.wflynn.cn/aitools/tre...
Following the WeChat Official Account "YourTools", GitHub's trend and open source tool is sent daily:
Auto-fetched and translated by AI Atlas. All rights belong to the original author.