Prompt = Subject/Background/Camera View... + Motion
1.Basic Structure: Since there is an existing scene, better to minimize (or even avoid) describing static/unchanging elements. When explicitly indicating moving subject, focus on describing motion, including:
Motion of the subject
Motion/changes in the background
Motion of the camera view (camera movement)
2.Be Concise: Use simple words and sentence structures. The model will expand on your prompt based on its understanding of the image to generate videos matching expectations.
3.Feature Description: If the subject has distinctive features (e.g., curtains, showerhead, TV), include them to better identify it. When describing motion, clearly specify intensity adverbs (e.g., quickly
, dramatically
).
4.Adhere to the Image: Prompts must be based on the input image content. Clearly state the subject and the desired action or camera movement.
Crucially, ensure prompts do not contradict the image content or basic parameters. Examples of contradictions:
Image shows a sofa, prompt says "a coffee table".
Background is a living room, prompt says "showerhead in bathroom turns on".
5.Negative Prompts Not Supported: The model does not respond to negative prompts.
Image | Prompt | AI-generated Video |
---|---|---|
![]() |
Toilet seat opening | ![]() |
![]() |
Curtains opening outward | ![]() |
![]() |
Make the starry ceiling move | ![]() |