Blocking

Blocking refers to the process of mapping out in advance where actors and cameras will move on stage before shooting a play or film. It's closer to a rehearsal where, instead of full costumes or lighting, you use simple markers in place of actors and figure out camera angles and movement first.

The same principle applies in AI video production. When you try to direct a fight scene or camera movement using only text prompts, limbs often get tangled or the frame jumps unexpectedly. To prevent this, creators first set the camera path and character movement inside a 3D program using simple shapes and boxes rather than human figures. This low-resolution animation is then fed into a video generation AI as reference material, and the AI follows that movement and camerawork to produce the final video.

In other words, instead of handing the AI a finished picture, you give it a sketch with arrows. The AI fills in the details, while humans lock down the underlying skeleton of movement.

In articles, this appears in phrases like "the movement blocked out in Blender is carried straight into Seedance 2.5." A common misconception is that blocking needs to look like a finished video, but in reality, textureless shapes and bounding boxes are enough. The key isn't detailed 3D modeling but capturing the skeleton of the movement.

Open a free 3D program (such as Blender) and use simple boxes or spheres instead of human figures to map out the movement you want. Then animate the camera's direction of movement alongside it. Upload this low-resolution animation file as reference material to a video generation AI service, add a prompt describing the scene you want, and compare how closely the final video follows the planned movement to get a feel for how it works.