Back to Discover

Kling says its new image AI makes sharper, more photo-like pictures with long, readable text

The opening shot of Kling AI's launch film: a tall red rock shaped like a head, in a desert at sunset, with dust blowing past it

Kling AI, the picture and video tool from Chinese social media company Kuaishou, has opened its newest image AI, released on Nov. 17, to all its users, though sharper pictures need a paid plan. Kling says it makes pictures that look closer to camera photos and can write long blocks of text, such as a shopping page's product description, into a single image. Kling's terms bar business use of the pictures without the company's written permission, though its paid plans allow it. The terms also let Kling train its AI on what people upload.

Kling AI, the picture and video tool from Chinese social media company Kuaishou, released Kling Image 4.0 on Nov. 17. It is Kling's newest model for making pictures from a written request or from pictures a user supplies. Kling says it is open to all its users, and to businesses through an API, a connection that lets developers plug a service into their own apps.

Kling's pitch is pictures that look real. It says the model shows skin, fabric, glass and metal, and the way sunlight fades into shadow, more like a photo straight from a camera. It can also make pictures at 8K, Kling's sharpest and most detailed size, which the company says keeps freckles and strands of hair sharp even when someone zooms in. Kling's plans page says making 4K pictures needs a paid plan, and it offers paid subscribers 8K pictures at a discount; it does not say whether free users can make 8K pictures. Kling says a single model both reads the request, including any pictures a user supplies, and draws the new image, which it says helps it place people and objects in a scene that hangs together.

The other pitch is words. Kling says the model can write long blocks of text into a picture, such as the product description on a shopping page or the labels on a diagram of a product's features, in languages including English, Spanish, Japanese and Chinese. Kling says one request can make a set of up to 10 pictures in a matching style, such as a film storyboard, or one long image laid out like a web page. Its earlier image model could already make sets. Users can also change the words in a finished picture on Kling's own site and apps.

Kling's terms come with conditions. They say users own what they make, but anyone who uses or shares the pictures must mark them as made by Kling AI unless Kling agrees otherwise. Business use needs the company's written permission, though Kling's plans page says paid subscribers may use the pictures for business, just not to build a rival product; the paid plans also remove Kling's watermark. The terms let Kling train its AI on what people upload, and show or reuse both uploads and finished pictures, including to promote its products. They tell people not to upload anything they consider confidential, and let users stop Kling using what they upload or make by emailing the company. They also warn that the pictures may not show real people, places or facts accurately.

The examples Kling has shown were made and picked by the company, so it is not yet clear how often an ordinary request turns out that well. Kling says it is still improving the model.

Sources

About this video

  • Available now: “Kling Image 4.0 is now available to all users.” Source
  • The official sources don't say whether the video is sped up or whether the footage is real, simulated or AI-generated.