Kuaishou Keling's New Features Image-to-Video and Video Continuation

TapTechNews on June 21, it was reported that the new features "Image-to-Video" and "Video Continuation" of Kuaishou Keling were launched today.

Image-to-Video function supports converting static images into 5-second videos, and users can control the movement of objects in the image through the prompt text; Video Continuation function supports one-click continuation and continuous multiple continuations of the generated video, and the longest can generate about 3 minutes of video; in addition, Text-to-Video added 9:16 and 1:1 video size options.

TapTechNews attached the official website of Kuaishou Keling: https://kling.kuaishou.com/

Keling is a large video-generation model independently developed by Kuaishou, which can generate large-scale reasonable movements and simulate the characteristics of the physical world.

Keling uses the DiT architecture, and Kuaishou has upgraded the latent space encoding/decoding, temporal modeling and other modules in the model.

In the latent space encoding/decoding, Kuaishou independently developed a 3DVAE network to achieve synchronous spatio-temporal compression and obtained a high reconstruction quality, achieving a balance between training performance and effect. In the temporal information modeling, Kuaishou designed an all-attention mechanism as a spatio-temporal modeling module.