GraphVid: Interactive Graph-Controllable Video Generation
GraphVid lets users steer multi-object video generation with structured interaction graphs instead of hand-drawn trajectories.
The paper says trajectory controls become hard to scale when scenes involve several objects, especially with occlusion or overlap. Its authors introduce a graph-conditioned image-to-video model and a new interaction-focused dataset, GraphVid-Bench, with relational annotations. They report stronger controllability and quality than Motion-I2V while using less training data and fewer trainable parameters. ArXiv · AI/CL/LG's note
The paper says trajectory controls become hard to scale when scenes involve several objects, especially with occlusion or overlap. Its authors introduce a graph-conditioned image-to-video model and a new interaction-focused dataset, GraphVid-Bench, with relational annotations. They report stronger controllability and quality than Motion-I2V while using less training data and fewer trainable parameters. ArXiv · AI/CL/LG's note
score 5