What changes in the model

Kuaishou has announced, through the outlet IT之家, the arrival of Kling Image 4.0, a new generation of Kling AI's image-creation model. The release, announced on November 17, is accompanied by the opening of API access to business customers and developers, an element that signals the intention to bring the model out of consumer use and into companies' production workflows.

The main novelty concerns the architecture: the model adopts an approach that Kuaishou describes as “unified understanding and generation,” with the goal of integrating visual understanding and image generation into a single system. According to the company, this link between interpreting the request and producing the visual content serves to make the path from creative intention to final result more direct.

The three stated directions

Kuaishou indicates three central upgrades. The first concerns realism and resolution: the model supports native 8K image generation for the first time and incorporates notions of physics and visual perception into the process, to make light, skin, and surfaces closer to a photographic shot. Such a high resolution has concrete utility for detailed visualization and for subsequent cropping in post-production.

The second concerns text within the image: Kling Image 4.0 claims the ability to generate up to 2,000 characters in a single image, with more accurate handling of complex instructions and layout, so as to keep even small characters legible. The third concerns content organization: the model can produce up to 10 images in a series from a single description, maintaining a shared style, and is also designed for long formats such as e-commerce detail pages and web layouts.

Input, languages and style control

The source material specifies some operational parameters: text input can reach 15,000 tokens, a threshold that should allow for detailed requests concerning text arrangement, layout, and subject presentation. More than 70 predefined commercial fonts are also integrated, along with the ability to upload reference images to guide the shape and style of the characters. The stated language coverage includes Simplified and Traditional Chinese, English, Japanese, Korean, Spanish, and Portuguese.

On the consistency front, the model has a capability described as “Agentic” that autonomously organizes the content and priority of the individual images in a series, from storyboard to product presentation. Finally, more than 110 predefined tonal styles are available, with the possibility of extracting colors and light from a reference image to build one's own visual profile.

To be verified

The information comes from the company's communications and IT之家's summary: there are no independent tests, benchmarks, or verifiable examples. It therefore remains to be established how native 8K actually performs compared with competing solutions, the actual accuracy of text at 2,000 characters, and the commercial terms of API access, on which the material provides no details.

Source and attribution: this original summary was prepared by AICronaca based on the report by IT之家 ↗. The statements are attributed to the source; any informational limitations: Content based exclusively on the announcement reported by IT之家 and on Kuaishou's statements: no independent verification, no data on prices, availability, benchmarks or API terms..