What it is
Kling is an AI video generator made by Kuaishou Technology, a Beijing company best known for Kuaishou, a short-video app that MIT Technology Review has described as the most prominent rival in China to Douyin, the Chinese sister app of TikTok. You describe a scene in words, or upload a still picture, and Kling makes a short video clip from it. The clip has never been filmed. Every frame is generated.
It does two main jobs:
- Text to video. You type something like "a paper boat drifting down a rain-soaked city gutter, low camera angle, evening light", and it produces a clip of that scene.
- Image to video. You upload a picture, such as a photo, a drawing or an image made with a tool like Midjourney, and describe how it should move. The picture becomes the first frame and Kling animates it. Many creators find this the more useful of the two, because the picture fixes how things look and the prompt only has to describe the motion.
Newer versions do more. At the time of writing, Kling can also create sound with the pictures, including speech, sound effects and background noise; make a clip that cuts between several camera shots; copy the movements from a video you upload onto a character in your picture; and edit an existing clip from a text instruction. It also makes still images.
The name is the English brand for the Chinese 可灵 (Kělíng). The international service runs at kling.ai (the older address, klingai.com, now forwards there). You use it in a web browser or through the Kling app, and developers can connect their own software to it through an API, a way for one program to use another. Kling models are also offered inside other companies' products, including Adobe's Firefly app and Runway, as the examples below explain.
How does it work? Kuaishou has said Kling uses a "diffusion transformer". That combines two ideas explained on their own pages. A diffusion model starts from random visual static and cleans it up, step by step, into pictures that match the prompt. A transformer, the architecture behind chatbots like ChatGPT, handles the video as small patches across space and time, so the whole clip is worked on together rather than one frame after another. This was the same general approach OpenAI described for Sora.
A short history, with dates mostly from Kuaishou's and Kling's own announcements:
- June 2024. Kuaishou unveiled Kling on 6 June 2024, less than four months after OpenAI's first Sora preview. Unlike Sora, which only a small group of testers could use at the time, Kling opened to a waiting list of users in China through Kuaishou's video-editing app.
- July 2024. Kuaishou opened Kling to users worldwide on 24 July 2024, with sign-up by email address. Its announcement said users got 66 free credits a day, enough for about six standard videos.
- Late 2024. Kling 1.5 (September) added full HD video and a "motion brush" for painting the paths that objects should follow. Kling 1.6 followed in December.
- 2025. Kling 2.0 launched on 15 April 2025, when Kuaishou said the service had passed 22 million users. Version 2.1 followed in May. In December, Kling 2.6 added sound generated together with the pictures, and a model called Kling O1 put generation and editing into one system.
- 2026. The Kling 3.0 series launched on 5 February 2026, with clips of up to 15 seconds, several shots in one clip, and speech in several languages. Kuaishou later added native 4K video (very high resolution) and a faster Kling 3.0 Turbo model.
Versions, lengths and features change every few months, so check the current ones on the site itself.
Why it matters
It showed that the frontier of AI video was not one company's. When Sora was previewed in February 2024, many people assumed OpenAI would have the field to itself for a while. Kling was usable by the public months before Sora was, and it was measured against Sora from then on. Sora itself was discontinued in 2026, as its own page describes, while Kling keeps shipping new versions. For anyone choosing tools, the lesson is that the field moves quickly and no single company owns it.
It is one of the first AI video tools to earn serious money. Most AI video products are expensive to run and it is often unclear whether they make a profit. Kuaishou reports Kling's revenue in its quarterly results. It said Kling's revenue passed 850 million yuan in the second quarter of 2026, more than three times the figure a year earlier, and that Kling's monthly revenue passed 20 million US dollars in December 2025. These are the company's own figures, but they come from its results announcements to investors, not from marketing copy. The same results also showed the cost: Kuaishou's overall operating profit fell that quarter as it spent heavily on AI.
It brings Chinese rules on AI content into creators' hands worldwide. AI services in China must follow government rules on content, including rules that require generated material to reflect "core socialist values". The DeepSeek page explains that wider picture, including US controls on the export of AI chips to China. For Kling this has had visible effects: testers have found that it refuses some politically sensitive prompts. It isn't only a Chinese question, though. Every video tool refuses something, and the reader deserves to know what each one blocks and why.
It makes moving pictures cheap for small creators. An author, a small shop or a teacher who could never pay for a film crew can make a few seconds of convincing footage for the price of a monthly subscription. That opens new ways to advertise and tell stories. It also makes convincing fake footage cheap, which is the other side of the same coin.
Real-world examples
Launch, a global waiting list, and what testers found (2024)
When Kuaishou announced Kling on 6 June 2024, it said the model could make videos of up to two minutes in full HD. Only a limited group of users in China could try it at first. Zeyi Yang of MIT Technology Review got in, which took a Chinese phone number, a waiting list and an extra application form. He reported on 19 June 2024 that the clips available to him were only five seconds long, at a lower resolution, and could look "blurry and grainy". A Chinese AI artist he interviewed said Kling was weaker on composition and colour but good at simulating physics and nature. A professor he spoke to warned that longer clips were more likely to go wrong, so five-second clips suited a short-video platform better than a film studio.
Kuaishou opened the service to the world on 24 July 2024. The same day, TechCrunch's Kyle Wiggers tested what it would and wouldn't make. Prompts such as "Democracy in China", "Chinese President Xi Jinping walking down the street" and "Tiananmen Square protests" returned a nonspecific error message. He also reported that the filtering seemed to work on the words of the prompt: Kling would animate a portrait of Xi Jinping if the prompt didn't name him. TechCrunch linked this to China's internet regulator, which tests AI models' answers on sensitive topics before they are released. Kling's rules and filters have changed since then, but the pattern is similar to what testers found with Chinese chatbots, covered on the DeepSeek, Qwen and Kimi pages.
Famous directors, and a critic's verdict (2024 to 2026)
On 9 September 2024, Kuaishou announced a "director co-creation" programme. Nine Chinese directors, among them Jia Zhangke, a filmmaker who has won major prizes at Cannes and Venice, would each make a short film of about three minutes, with every shot generated by Kling. It was Jia's first AI film. The shorts were released in late 2024, and Kuaishou used them to show what the model could do for professional filmmaking.
Not everyone was convinced. Writing in the Los Angeles Review of Books in April 2026, Michael Berry, a scholar of Chinese film, noted that Kling had sponsored Jia's short, Wheat Harvest, and that it effectively served as an advert for the brand. He found it "utterly plastic, mechanical, and soulless", a strange result from a director famous for gritty realism. Kuaishou, for its part, reports prizes for work made with Kling, including one Silver and two Bronze Lions at the 2026 Cannes Lions advertising festival for two adverts made with it. The two views can both be true. The tool can produce award-winning footage, and a sponsored showcase can still say more about the sponsor than about the art.
Fake Kling websites spreading malware (2025)
Popularity attracts criminals. On 20 May 2025, the security company Check Point reported a campaign of around 70 sponsored Facebook posts from fake pages pretending to be Kling. They led to copycat websites that invited visitors to make an AI image or video. Instead of a video, the visitor got a download dressed up as a picture, with a name like "Generated_Image_2025.jpg", which actually installed software that let attackers take control of the computer and steal data. The lesson applies to every popular AI tool: type the address yourself, and be suspicious of any "AI generator" that asks you to download and open a file to see your result.
Kling inside Adobe and Runway (2025 to 2026)
Kling isn't only used on Kling's own website. Adobe began adding outside "partner models" to its Firefly app in 2025, and at the time of writing its help pages list Kling 2.5 Turbo, Kling 3.0 and Kling 3.0 Omni. Adobe's page includes a telling note: "Chinese models like Kling" are offered only to individual subscribers, not to teams or business plans. Adobe also says it does not make its commercially safe promise for partner models, as the Adobe Firefly page explains. Runway offers Kling 3.0 on its paid plans, alongside its own models. For a creator, this means you may already be using Kling without visiting kling.ai, under a different company's prices and terms.
An everyday scenario: a picture-book author's moving illustration
Picture Maya, who has self-published a children's picture book about a fox who can't sleep. She wants a ten-second clip for her social media. She signs in to Kling with an email address, uploads her own illustration of the fox curled under a tree, and writes a prompt describing only the motion: "the fox slowly opens one eye, leaves drift down, the camera pushes in gently". The first result is lovely except that the fox's face changes shape halfway through and grows an extra ear. She tries again with a shorter, calmer prompt and gets a usable clip on the third attempt. Because she is on the free plan, the clip carries a Kling watermark and isn't licensed for commercial use, so she posts it as a "behind the scenes" teaser with a note that it was animated with AI, and decides to take a month's paid plan before she uses anything like it in an advert for the book.
Try it yourself
Menus, prices and free allowances change often. Treat these steps as a guide, and check the current terms on kling.ai before you rely on anything here.
- Go to kling.ai and type the address yourself. Don't follow an advert or a link in a social media post. As the Check Point example shows, fake Kling sites exist.
- Create an account. At the time of writing you need one to generate anything. Kling offers sign-in with Google, Apple or an email address, and the terms set a minimum age of 13 (higher where local law says so), with a parent or guardian involved for minors.
- Check what's free. At the time of writing, Kling's pricing page describes a monthly allowance of free credits for people who don't subscribe, claimed when you log in, and paid plans starting at around 10 US dollars a month or less, with first-month discounts. Watermark removal, 4K video and commercial use are listed as paid-plan benefits. Credits are used up faster by longer clips, higher resolution and newer models, so start small.
- Start from a picture you own. Choose image to video and upload something that is yours: a photo you took, your own drawing, or an image you made with an image tool whose terms let you use it. Avoid photos of real people who haven't agreed.
- Describe only the motion. The picture already shows what things look like. Try a prompt like this, adapted to your image: > Slow, gentle camera push-in. Light wind moves the leaves and the character's hair. The character blinks and turns their head slightly towards the camera. Calm, quiet mood. No text on screen.
- Generate, then look closely. Watch the clip three times. Look at hands, faces, eyes, text and anything that should stay solid. Note which parts of your prompt it followed and which it ignored.
- Change one thing at a time. If something went wrong, simplify. Shorter prompts with one clear movement usually work better than a long list of actions. Try the same image with two different prompts and compare.
- Try text to video for contrast. Write a prompt from scratch with the subject, setting, camera and mood spelled out, and see how much less control you have than when you started from a picture.
- Optional: test a refusal. Ask for something the rules forbid, such as a realistic video of a named politician, and see what happens. Every video tool blocks some things. Knowing what yours blocks helps you plan.
- Read the terms before you publish. See "Common mistakes" below for what Kling's terms say about commercial use, labelling and your data.
Common mistakes and misunderstandings
Assuming the free plan covers business use. At the time of writing, Kling's pricing page says paid subscribers may use what they generate for commercial purposes, and lists this among paid benefits. Its general terms say that, without Kling's written permission, you may not use the output commercially, and that you should label output as made by Kling. If you plan to sell a video or use it in an advert, check the current terms and expect to need a paid plan. The same applies inside Adobe or Runway, where their own terms apply too.
Not reading what happens to your uploads. At the time of writing, the international service is run by Kling AI Pte. Ltd., a Singapore company, and its privacy policy says your data is stored on servers in Singapore and may be shared with companies in the same corporate group. Its terms give Kling a licence to use what you upload and generate, including to train its AI models, and say you can withdraw that permission by emailing support. Don't upload anything private, such as photos of your children or unreleased artwork, without reading these first. The Privacy page explains why this matters for every AI tool.
Believing the demo reel. Company showcases are chosen from many attempts. Testers regularly report faces that shift, limbs that multiply and objects that melt, especially in longer clips. A review of Kling 3.0 by the AI film school Curious Refuge in February 2026 praised its physics and camera control but found consistency "collapsed" in 2D animation, and judged that it put physics ahead of following the prompt. Expect to generate several versions and keep the best one.
Thinking it's neutral on every subject. As the TechCrunch test showed, Kling has refused politically sensitive prompts about China, which fits the rules Chinese AI services must follow. Western tools have their own blocks and biases too. If your project touches politics or history, know that the tool's rules may shape what you can make.
Assuming you own the copyright. Kling's terms say it doesn't claim ownership of what you make. That isn't the same as the law giving you a copyright. In the United States, at the time of writing, raw AI-generated material isn't protected by copyright on its own; the Midjourney page covers the rulings. Your own drawings, script and editing choices are a different matter.
Forgetting to label AI video. Realistic AI video can mislead people even when you didn't mean it to. Kling's own terms ask you to label its output, China's rules on labelling AI content took effect on 1 September 2025, and many platforms have their own rules. Say clearly when footage is AI-made, and never make realistic video of a real person without their permission.
Treating one tool as the only choice. Kling is one of several AI video tools, alongside Veo and Runway, and each company publishes comparisons that favour itself. Try the same image and prompt in two or three tools and judge the results for your own project.
Related elements
- Sora: OpenAI's video generator, previewed shortly before Kling appeared and discontinued in 2026.
- Runway: a video-generation company that also offers Kling models inside its app at the time of writing.
- Veo: Google's video model, another of Kling's main rivals.
- Adobe Firefly: Adobe's creative app, which offers Kling as a partner model to individual subscribers.
- Diffusion Model: the technique that turns random static into the frames of a Kling video.
- Transformer: the architecture Kling combines with diffusion to handle a clip as patches in space and time.
- Midjourney: a popular way to make the still images that many creators then animate in Kling, and the page that covers AI copyright rulings.
- DeepSeek: the Chinese AI lab whose page explains China's content rules and US chip export controls.
- Qwen and Kimi: other Chinese AI models, where testers have found similar political filtering.
- Privacy: what to think about before you upload photos and artwork to any AI service.
- World Models: the bigger idea that video generators like Kling might learn how the physical world behaves.