Currently there are multiple ways to write rotations for fewest-moves. This ultimately leads to more ambiguity (which is something we should try to minimise to improve FMC grading consistency) and higher chances of false DNFs (some FMC g...
I2V and V2V have strong control over the appearance of generated video, as separate images are usually provided to set the scene. As for the motion, I2V shows limited controllability since the task is defined to accept only coarse-grained motion clues, suc...