Hey gang, wanted to give a new update on my creative workflow regarding Seedance 2.5: I was over-controlling the scene before the video was even generated. I’d create reference images with the characters already positioned in the setting, holding the props, with the composition figured out. In my testing, that seemed to limit how naturally Seedance could interpret the action. Now I’m keeping the references separate: • Character sheets for identity and wardrobe. • Setting references for the environment. • Object references for the relevant props. Then I use the prompt to explain how everything comes together: what happens, what the characters feel, and the camera direction. That gives Seedance room to figure out the staging, movement, and performance. What I’ve realized is that the number of references isn’t necessarily the issue. It’s how much we’re forcing the scene through them. Useful references give it context. Leaving the composition open gives it creative freedom. So far, that balance is giving me more fluid, authentic results. I've included my workflow diagram of how I use the multi reference tool, and an example of one of our newest pieces that shows you what this model can really do when you allow it to open up and run with more control. These models are getting more powerful by the release and over controlling them can lead to lesser quality. You are the director, be smarter than the model but trust it that it can pull off the task of what it's being asked to do and it may just blow your mind