after nine iterations in three months, kuaishou keling ai released the 1.5 model to the world
2024-09-22
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
on september 19, keling ai had a major upgrade, adding the keling 1.5 model to video generation, which has significantly improved image quality, dynamic quality, aesthetic performance, motion rationality, and semantic understanding. at the same time, keling ai also introduced a new "motion brush" function to further improve the precise control capabilities of video generation.
first, the base model has been upgraded again, with the addition of the new keling 1.5 model, which supports direct output of 1080p hd video in high-quality mode, challenging the clarity and texture of large screens. compared with the keling 1.0 model, the 1.5 model has significantly improved picture quality, dynamic quality, text responsiveness, etc., and its overall effect has been improved by 95% in internal evaluation.
previously, the keling 1.0 model could generate 720p videos in high-quality mode. after this upgrade, the 1.5 model can directly generate 1080p high-definition videos in high-quality mode. by inputting the prompt word "girl looking at the car window" and comparing the effects of the two versions of the video, it can be found that the picture quality of the new keling 1.5 model has been significantly improved: the picture clarity is intuitive, the facial details of the girl on the right side of the picture are clearer and richer, and the water mist on the car window and the overall light and shadow performance are also better. at the same time, the overall composition of the picture is further optimized under the new model, and the picture is more beautiful.
the new model also has significant improvements in dynamic quality. for example, compared with the popular noodle-eating case of keling ai, the prompt word "little boy eating noodles" is input. in the video generated by the 1.5 model on the right, the noodles have very realistic physical performance in terms of elasticity and drape from being picked up to the mouth. at the same time, the boy's right hand holding chopsticks and chewing when eating noodles are more natural and smooth than the 1.0 model on the left, and the overall movement rationality is greatly enhanced.
in terms of image-generated videos, keling's new 1.5 model can respond to more complex text description requirements. for example, through a food photo without people and the prompt "zoom out, a little boy walks to the table, picks up a spoon and starts eating", in the generated video, as the camera shakes slightly, a spoon "enters the scene", and then the picture focuses on the little boy holding the spoon, watching him put a spoonful of food into his mouth, and the details of the spoon pushing the rice grains in the bowl are also presented in detail, showing a strong ability to understand image-generated videos.
in this upgrade, keling ai also brings a powerful "motion brush" function, which greatly improves the creator's ability to control the motion effect when generating videos from pictures. the "motion brush" function supports specifying motion trajectories for elements in the picture (people or objects, etc.). users only need to outline the part of the picture that needs to control the direction of movement, and then draw an arrow to indicate the direction of movement to achieve precise motion control. this function supports specifying motion trajectories for up to 6 elements (people or objects, etc.) in the picture after uploading the picture. in addition, you can also specify additional static areas for certain elements to give the video content better motion control and motion performance.
currently, pictures of various sizes (16:9, 4:3), 9:16, 3:4, and 1:1) can all be generated using the "motion brush" to generate videos, with a duration of 5 seconds. a large number of user and media reviews show that the motion brush function of keling ai is industry-leading in terms of ease of use and performance.
in fact, keling ai has also recently upgraded a series of other functions, such as supporting the generation of up to 4 videos at a time, which allows creators to quickly select the best generation result; the "image to video" function now supports 10-second duration and supports adding tail frames in standard mode; the "ai picture" function supports "image quality enhancement". in addition, the official user guide has been launched to help users better control keling ai.
since its release in june this year, this is the 9th iteration and upgrade of keling ai. this upgrade will also be launched globally. in july this year, keling ai announced that the international version 1.0 was officially launched and officially opened to global users. subsequently, a global membership system was launched. at present, keling ai has accumulated a large number of domestic and foreign users. gai kun, senior vice president of kuaishou and head of the main site business and community science line, previously disclosed at the kuaishou investor day that more than 2.6 million people have used keling ai and generated more than 27 million videos and 53 million pictures. (author: liu jia)
