Seedance 2.5 prompts: a checklist for 30-second clips

The checklist is aimed at a shift in how prompting works once a clip gets longer and carries its own sound. With shorter, one-shot clips, a writer can describe a single image and let the model fill in the rest. At around 30 seconds, the checklist argues, the composition has to be planned by the second — for example, a hook, then motion, then a benefit, then a call to action inside one clip — so that the intended sequence does not collapse.

The checklist also treats reference assets as something to assign a job to, not simply to accumulate. Images, video and audio can each be tagged for what they should hold fixed, whether that is a product's appearance, a person's face and outfit, a colour and camera mood, a motion reference or an audio tone. That matters most, the checklist says, when the same person or product has to appear across several scenes.

On audio, the checklist notes that Seedance 2.5 generates sound and video at the same time, so dialogue, music, ambience, sound effects, rhythm and emotional tone belong in the prompt itself, with mouth-movement synchronization called out where there is dialogue. If the sound will be replaced later, the checklist suggests narrowing the request to ambience only from the start.