Describe the scene, timeline, camera, and audio together
The official workflow guidance recommends stating the overall scene first, then breaking the action into timed shots. Camera movement and audio—dialogue, sound effects, and music—belong in the same prompt block.
- Start with location, subjects, and the overall action.
- Break temporal changes into explicit shots or beats.
- Describe camera motion and accompanying audio alongside each shot.
Explain what should happen between the first and last frame
For I2V or FL2VA, the connected images establish optional first and last frames. The prompt should describe the intended shots, motion, and audio rather than merely repeating what is visible in the images.
Give every reference a specific job
For Ref2VA workflows, refer to inputs by their ordered tags, such as <Picture 1>, <Video 1>, and <Audio 1>. State whether each reference controls identity, style, motion, camera behavior, or voice.
- Keep reference labels in the same order as the connected inputs.
- Avoid ambiguous instructions such as “use this reference” without assigning a role.
- Use the Ref2VA weights for reference-driven generation; they are distinct from FL2VA.