Kling AI is a fast-moving video and image generation service from a Hong Kong listed Chinese company, and teams adopting it need clear facts on versions, access routes, pricing and data terms before they start. Kling AI is a generative video and image platform from Kuaishou Technology, first released for beta testing in China on 10 June 2024, with its current flagship, Kling 4.0, launched on 15 October 2026. It produces video and images from text, images, video and audio references, through a web studio, a developer API and agent tooling (MCP and CLI). This page sets out the version history, what each generation added, how to access Kling, what is published about pricing, and the checks to run on provenance and data handling. It is written for creators, developers, and the security, privacy and procurement teams who approve AI tools.
Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on Kling AI's public documentation and Kuaishou Technology's press releases as of November 2026; it is not sponsored or endorsed by Kuaishou or any other provider. Product features and prices change, so confirm them against Kling AI's current documentation.
Key takeaways
- Made by Kuaishou Technology. Kling's press releases are issued by Kuaishou Technology, listed in Hong Kong under stock codes 01024 (HKD) and 81024 (RMB).
- Kling 4.0 is the current flagship. Launched on 15 October 2026, it generates up to 30 seconds in a single pass, with up to 10 keyframes and 10-bit HDR output announced.
- The 3.0 series added native audio. Announced on 5 February 2026, Video 3.0 generates speech in English, Chinese, Japanese, Korean and Spanish, in clips of up to 15 seconds.
- Several access routes. The Kling AI studio, the Kling AI Developer Platform API, Kling MCP and CLI for agents, and third-party platforms that resell Kling models.
- Check terms before upload. Read the Kling terms of service on ownership, training use and governing law, and settle anything unclear in contract before confidential material is used.
What is Kling AI?
Kling (in Chinese, 可灵) began as Kuaishou's self-developed video generation model. Kuaishou's June 2024 launch release describes it as built on a diffusion transformer (DiT) architecture with a self-developed 3D VAE for latent-space encoding and a full-attention spatiotemporal modelling mechanism. Today Kling AI describes itself as a platform to "Create high-quality AI videos and images with Kling AI" in one studio, covering text to video, image to video, an AI video editor, text to image, image to image, an AI photo editor, Motion Control and AI Avatar. It also offers Canvas, a node-based workspace where an agent helps brainstorm, write scripts and build storyboards.
By February 2026 Kuaishou reported more than 60 million creators worldwide, more than 600 million videos produced and more than 30,000 enterprise clients since launch, figures stated in the Kling 3.0 release.
Key facts
As of 15 November 2026.
| Field | Value | Source |
|---|---|---|
| Provider | Kuaishou Technology (Kling AI) | Kuaishou press releases |
| First release | 10 June 2024, beta test in the KuaiYing app in China | Kuaishou launch release |
| Current flagship | Kling 4.0, launched 15 October 2026 | Kuaishou Kling 4.0 release |
| Other current models | Video 3.0, Video 3.0 Omni, Image 3.0, Image 3.0 Omni, Kling Image 3.0 | kling.ai; Kling 3.0 release |
| Modalities | Video with native audio, images; text, image, audio and video inputs | Kling 3.0 release |
| Maximum single-pass length | 30 seconds (Kling 4.0) | Kling 4.0 release |
| Access routes | Web studio, Developer Platform API, MCP and CLI | kling.ai; kling.ai/dev |
| Open weights | No (hosted service) | kling.ai |
| Provenance labelling | Not stated by Kuaishou on the pages reviewed | Not applicable |
| User figures | 60 million+ creators, 600 million+ videos, 30,000+ enterprise clients (February 2026) | Kling 3.0 release |
Models and versions
Kling 4.0
- Length: native single-pass generation up to 30 seconds; Kuaishou says video extension will "soon support multiple extensions of up to two minutes".
- Picture: 1080p, with 10-bit HDR, 4K and 21:9 ultrawide described as coming.
- Control: up to 10 keyframe inputs and Omni Reference with up to 15 references, drawing on images, videos, subjects and voices.
- Prompts: up to 8,000 tokens of input.
- Agents: already integrated with Kling MCP and CLI for Omni Reference and keyframe-to-video generation.
- API: the release says capabilities "will soon be made available to enterprise customers and developers through API access".
The release's Omni Reference sub-limits (10 images, 5 videos, 7 subjects) add up to more than the overall 15, so treat 15 as the ceiling and test the combination you need.
Kling 3.0 series
- Audio: native speech in English, Chinese, Japanese, Korean and Spanish, plus English accents and Chinese dialects; multi-character dialogue where each character can speak a different language.
- Storytelling: multi-shot generation with shot-reverse-shot and voice-over; a storyboard feature in Video 3.0 Omni sets duration, shot size, perspective and camera movement per shot.
- References: Video 3.0 Omni can extract visual traits and voice characteristics from an uploaded reference video.
- Images: Image 3.0 models output 2K and 4K, and can keep or generate text such as signage and logos.
- Inputs: text, images, audio and video across text to video, image to video, reference to video and in-video editing.
Pricing and plans
Kling publishes API pricing on its Developer Platform at kling.ai/dev/pricing and sells memberships inside the studio, including an Ultra tier named in the Kling 3.0 release. Membership prices and credit allowances are not stated in the public pages reviewed for this guide, so take them from the studio at the time of purchase. New models have reached top-tier members first: the 3.0 series went to Ultra subscribers before the public.
Kling models are also resold through third-party platforms at their own published rates. As one dated example from a reseller's official API price list (Runway, $0.01 per credit, 15 November 2026):
| Model id on the reseller API | Resolution or mode | Credits per second | USD per second |
|---|---|---|---|
| kling_4 | 1080p | 23.5 list (18.8 with a discount to 30 November 2026) | $0.235 list |
| kling_4 | 720p | 15.7 list (12.56 discounted) | $0.157 list |
| kling_4_flash | 720p | 9.4 (29-credit minimum) | $0.094 |
| kling3.0_pro | With audio | 17 | $0.17 |
| kling3.0_4k | 4K | 43 | $0.43 |
| kling2.5_turbo_pro | Standard | 12 | $0.12 |
Reseller rates are set by the reseller and are not Kling's own prices. Compare the total cost per finished clip, including retries, rather than headline rates.
Features
- Video: text to video, image to video, reference to video, Motion Control, keyframes and an AI video editor.
- Image: text to image, image to image, native 4K image generation, series images with consistent style, AI photo editor.
- Audio: native multilingual speech in the 3.0 series; voices as Omni Reference inputs in 4.0.
- Characters: AI Avatar and Element Library 3.0.
- Workflow: Canvas for agent-assisted scripting and storyboards; MCP and CLI for automation.
Availability
| Route | What it is | Typical user |
|---|---|---|
| Kling AI studio | Web studio at kling.ai for video, image, editing, Canvas and avatars | Creators and marketing teams |
| Kling AI Developer Platform | API at kling.ai/dev, with pricing and documentation pages | Developers building products |
| Kling MCP and CLI | Agent-driven batch generation and multi-task production | Teams automating with AI agents |
| KuaiYing | Kuaishou's video editing app, the original beta route in China | Users in China |
| Third-party platforms | Other services that resell Kling models under their own accounts and terms | Teams already on those platforms |
Each route has its own account, credentials and terms. An approval for one route does not cover the others: a Kling model used through a reseller is governed by the reseller's terms and data handling as well as Kling's.
API and MCP credentials are non-human identities that can spend prepaid balance and generate content in the organisation's name. Keep them in a secrets manager, issue one per application and rotate them. See the AI Model Provider API Keys Leak Response Guide.
Provenance and labelling
Kuaishou's pages reviewed for this guide do not describe C2PA Content Credentials or an invisible watermark for Kling output. Teams publishing Kling video should therefore label AI-generated content themselves where law or policy requires it, and keep originals with whatever metadata the export carries.
- China: generative AI services offered in China are subject to the national AI content labelling rules; see the China Generative AI Measures and AI Labelling Rules Guide.
- EU: deployers who publish deepfakes have disclosure duties under Article 50; see the EU AI Act Article 50 Transparency Obligations Guide.
- Standards: for how Content Credentials work, see the AI Content Provenance and C2PA Guide.
Likeness and voice
Kling 3.0 Omni can take visual traits and voice characteristics from an uploaded reference video, and Kling 4.0 accepts voices and subjects as references. These features make it straightforward to reproduce a real person's appearance and voice, so organisations should require documented consent before any real person's likeness or voice is used, and keep that record with the project. See the Deepfake and AI Impersonation Guide.
Data and privacy
- Terms: Kling's terms of service are at kling.ai/docs/terms-of-service; read the sections on output ownership, commercial use, training use of inputs, minimum age and governing law before rollout.
- Enterprise: the Developer Platform describes its tools as secure and enterprise-ready; ask for security attestations, data residency, retention periods and a subprocessor list in writing.
- Cross-border data: Kuaishou is a Chinese company; assess where prompts, uploads and outputs are processed, and see the China Cross-Border Data Transfers guide where data flows involve China.
- Staff use: set rules on which data classes may be uploaded, using the AI Acceptable Use Policy Template.
Release history
Dates and changes below are as stated by Kuaishou in its Kling releases.
| Date | Version | What changed (as stated by Kuaishou) |
|---|---|---|
| 10 June 2024 | Kling (first model) | Text to video up to two minutes, 30fps, up to 1080p, multiple aspect ratios; beta in KuaiYing for users in China |
| Before February 2026 | Kling 2.x series (including 2.5 Turbo and 2.6) and Kling O1 | Named as the foundation for the 3.0 series; individual release dates are not given in the releases reviewed |
| 5 February 2026 | Video 3.0, Video 3.0 Omni, Image 3.0, Image 3.0 Omni | Up to 15 seconds, native multilingual audio, multi-shot storytelling, 2K and 4K images, early access for Ultra subscribers |
| 15 October 2026 | Kling 4.0 | Up to 30 seconds single pass, up to 10 keyframes, 10-bit HDR announced, Omni Reference inputs, 8,000-token prompts |
Limits
- Announced versus available: Kling 4.0's 4K, 21:9 and two-minute extension features were described as coming at launch; test what your account can do today.
- Staged access: new models have reached top-tier subscribers first, so a plan may not include the newest version immediately.
- Reference limits: Omni Reference caps total references at 15, with overlapping sub-limits.
- Price drift: API and reseller rates change, and promotions carry end dates.
Practitioner checklist
- Record each Kling access route in use (studio, API, MCP, reseller) in the AI system inventory with owner and approved use cases.
- Read the Kling terms of service and, for confidential material, agree training use, retention and processing location in contract.
- Keep API and MCP credentials in a secrets manager, one per application, with alerts on balance and usage spikes.
- Require written consent before any real person's face or voice is used as a reference.
- Label AI-generated video in line with EU, China and other applicable rules, and keep originals.
- Re-check model availability and prices on kling.ai before budgeting, as features and promotions change.
- Review any reseller's own terms separately when Kling models are used through another platform.
Standards and references
- Kling AI: Next-Gen AI Video and Image Generator
- Kling AI: Developer Platform
- Kuaishou Technology: Kling AI Officially Launches Next-Generation Kling 4.0 (press release via GlobeNewswire)
- Kuaishou Technology: Kling AI Launches 3.0 Model (GlobeNewswire)
- Kuaishou Technology: Kuaishou Unveils Proprietary Video Generation Model Kling (PR Newswire)
- Runway API: Pricing (reseller rates for Kling models)
Related NHI Mgmt Group resources: AI Image, Video, Voice and Music Models Hub · AI Content Provenance and C2PA Guide · AI Model Provenance and Watermarking Guide · Shadow AI Policy Template