From https://x.com/Adam38363368936

Many people find that the randomness when generating videos with Seedance2.0 is extremely high, which is very frustrating.

Today I’ll share a prompt technique for Seedance2.0: concretize all the elements involved. For example, use specific reference images to define the character, clothing, scene, and actions. This can prevent the generated video from going off track. The richer and more detailed your definitions are, the more stable the generation will be.

Seedance2.0 prompt  The camera is pointed at a pair of jade-white feet, the lens slowly circles around the character in 【@图1】, gradually panning upward, arriving at the character’s hips, then finally moving to the character’s chin, without showing the full face. The camera then pulls back to reveal the character’s entire upper body. Inside the scene from 【@图2】, the character begins to sway and twist her graceful, alluring figure in a dance, performing a Western Regions style dance.