Choose Seedance 2.5 for longer narratives and editing workflows. Use the Seedance 2.0 variants when a shorter task calls for a different resolution or iteration path.
Seedance 2.5 text to video
Start with a clear scene brief. Name the subject, the action, the setting, the time of day, and the camera move you want to see. A short prompt with concrete direction is easier to revise than a paragraph that tries to describe every possible detail at once.
Seedance 2.5 image to video
Use an image when the opening look matters. Describe what should move, what should remain stable, and how the camera should travel through the shot. A product photograph can become a short motion study; a character image can set the visual starting point for a scene.
Seedance 2.5 reference control
Seedance 2.5 accepts image, video, and audio references in the supported workflow. Give each asset a purpose: an image for appearance, a video for pacing or composition, and audio for the sound direction. The official model description calls out more precise understanding of reference videos, including intent, framing, and cinematic language.
Seedance 2.5 audio-video generation
The model is designed for joint audio-video generation rather than treating sound as an afterthought. When sound matters, describe the environment and timing in the prompt, then review the output as a whole. Keep dialogue, music, and effects within the scope your provider supports for the chosen request.
Seedance 2.5 30-second storytelling
Seedance 2.5 is built for videos up to 30 seconds in one generation. That extra room suits a small narrative arc: establish the setting, show the action, then land on a closing beat. The official page also describes two extension opportunities when a longer sequence is needed.
Seedance 2.5 editing and extension
Use editing when a clip already has useful material and only one part needs to change. Use extension when the ending needs another beat. Frame-based and editing operations may require adaptive settings, so check the controls before sending the request instead of assuming every mode accepts the same options.
