Seedance 2.5's biggest strength is its multimodal reference system. Instead of describing everything in words, you attach real materials and let the model animate around them. This guide walks through each input type and how to combine them effectively.
Reference images are the workhorse. Attach photos of your character, product or scene and Seedance 2.5 will preserve their appearance across the whole clip. A practical pattern: one clean front-facing shot of the subject plus one environmental reference for style and lighting. In the prompt, describe the action — "the woman in the reference image walks through the neon-lit street" — rather than re-describing what the subject looks like.
Reference videos teach the model motion. A short clip of a dance move, a camera pan or a fabric simulation gives Seedance 2.5 a concrete motion target. Keep reference videos short and focused; the model transfers their movement style onto your subject rather than copying them frame by frame.
Reference audio drives sound. Attach a voice sample and the generated character can speak with a similar tone and pacing. This is what makes Seedance 2.5 genuinely useful for short drama: an actor reference image plus a voice reference equals a consistent, speaking character.
First and last frame control is the precision tool. Upload the exact opening frame and closing frame, and Seedance 2.5 generates the in-between journey. Combine a first frame with a reference image of a different outfit and you can show a character changing clothes within one continuous shot.
A few practical limits to remember: reference inputs are mutually exclusive with first/last frame inputs, so choose one workflow per request. For billing, clips that include reference video add the input video duration to the charged length.
Try the full reference workflow yourself in the Seedance 2.5 studio — no installation required.