Tencent Releases SCoPE

Tencent ARC released SCoPE (Sightline-Coordinate Positional Encoding). This model adds camera sightlines as positional coordinates to a pretrained video diffusion transformer, generating videos according to desired camera movements given the first frame, text prompt, and camera trajectory as inputs.

It is characterized by the ability to precisely control camera motion while preserving prior knowledge from existing image-to-video (I2V) conversion. This release is a self-contained package based on Wan2.2-I2V-A14B, including all elements required for inference without the need for separate checkpoint downloads.