Use text-to-video when you want to invent a scene without a starting image. Describe the subject, setting, action, camera movement, and sound. This is a useful starting point for exploring a new visual idea before preparing additional material.
Use image-to-video when you already have a frame that establishes the subject and composition. Write the motion you want to add: how the subject moves, how the camera responds, and what changes in the environment. A clear starting image and a focused movement instruction are easier to assess than a brief that changes the whole scene at once.
Choose reference mode when you need to combine material for appearance, movement, or sound. These inputs give you more ways to explain the intended shot; they do not replace the need to describe how the scene should unfold.
Use reference mode to combine appearance, motion and sound. The unified Seedance 2.5 adapter allows up to 30 image, 10 video and 10 audio references, 4–30 second output and 480p/720p/1080p. Video reference length contributes to the quote. For a longer story, generate separate shots and edit them together; it is not a single unlimited-duration generation.