Updream Launches Previs Studio: AI Video Creation Enters a 'Rehearsal Era'

The core problem with current AI video tools, the report explains, is the lack of controllability over 3D space and camera movement. Text prompts can say "camera slowly moves forward" or "a character crosses from the left," but they cannot precisely specify speed, distance, or final framing. This leads to repeated trial and error, commonly known as "card drawing." By using a rough 3D blockout as a spatial anchor, the model can better understand depth, occlusion, proportion, and movement space. This idea echoes the earlier use of ControlNet in the Stable Diffusion era, where depth maps or simple 3D models helped stabilize image generation.

Updream's Previs Studio lowers the barrier to creating such blockouts. Instead of requiring months of learning Blender or Maya, users can upload one to three reference images, and the tool generates a 3D white-box scene in about four to seven minutes. The scene can be rotated, scaled, and navigated. Users can then add characters, position cameras, draw movement paths, and record a preview video to be used as a reference for the final AI-generated video.

The report describes tests of three difficult scenarios. In a one-shot continuous shot, a woman in a cheongsam walks to a seat, and the blockout reference kept the background stable and the movement path consistent. In a multi-camera sequence, the blockout helped maintain character identity, spatial direction, and environmental continuity across cuts. In a complex camera movement inside a subway car, the reference improved execution stability and reduced spatial distortions. One tester noted that the white-box reference reduced the number of invalid attempts, especially for complex spatial scheduling.

However, the tool is not a one-click solution. The quality of the blockout depends on the input reference images, and very complex movements still require patient fine-tuning. The final rendered video may still have issues with character consistency or physics details, which need post-processing. But the shift is significant: creators can now decide how they want to shoot before generating, rather than guessing how the model will interpret a prompt.

As AI video models have matured, the industry's competitive focus is moving from whether a model can generate a video to whether a creator can control the result. According to the report, Updream is currently running a "Previs Studio Challenge" with up to 10,000 yuan in rewards, encouraging users to create complex shots using the new workflow.