快手可灵 How do you write prompts for Kuaishou Kling's text-to-video? What techniques are there?

First, let me share several sets of video images generated from prompts and how they compare with Sora. To use Kling well, you need a deep understanding of visual composition; only then can you fully describe the visual elements contained in the prompt. The more detailed it is, the closer you will get to your expected result.









The prompts are used in the same way as with previous AIGC generation tools. They use a relatively detailed visual structure, including people/objects, scene, style, color, climate, generation medium, etc. That is, they include both the element framework for image creation and spatiotemporal elements such as climate and location, reflecting the characteristics of a large model that understands the physical world.



As can be seen from the demo content, although Kling’s results are still slightly worse than Sora’s, its advantages lie in saving compute, being fast, and having no usage barrier. Kling and Sora are products of the same dimension. Previously, Sora was a dimensionality-reduction strike against other production tools, but the emergence of Kling makes Sora seem less advantageous. Douyin and Kuaishou are both continuously launching powerful AI tools; this time they have won plenty of face for Chinese people.