seedance-prompt

Structure the prompt as Subject + Action + Scene + Camera + Lighting/Style + Audio + Constraints, leading with the subject and one visible action. Keep one main camera move, physically motivated lighting, and explicit preservation constraints, then remove filler to fit the prompt budget.

T2V prompts must build the entire shot from scratch: subject, action, scene, camera, light, style, and constraints. I2V prompts start from a reference image and should only describe motion, camera, timing, audio, and preservation rules, avoiding re-describing what the image already shows.

Yes, R2V and V2V modes accept reference assets, but each asset needs an assigned role such as identity, pose, scene, or style. State explicitly what must not transfer, and use only owned, licensed, or authorized references to avoid copying likenesses.

Generic results usually come from vague quality boosters, too many events in one clip, or emotion words instead of visible gestures. Keep one visible beat per clip, write performance as physical action, and cut duplicate style adjectives and speculative labels.

Yes, prompts can be written or compressed in Chinese, Japanese, Korean, Spanish, and Russian, including mixed-language versions. Language mixing should be used only for clarity in reference roles, dialogue, and technical camera terms.

Use FLF2V when you have both a starting image and a target ending image, since it treats the second image as the explicit final visual target. Standard I2V is better when only the starting frame is fixed and the motion outcome is open.