Translations:Diffusion Models Are Real-Time Game Engines/56/en

Video Quality. We use the auto-regressive setup described in Section 2, where we iteratively sample frames following the sequences of actions defined by the ground-truth trajectory, while conditioning the model on its own past predictions. When sampled auto-regressively, the predicted and ground-truth trajectories often diverge after a few steps, mostly due to the accumulation of small amounts of different movement velocities between frames in each trajectory. For that reason, per-frame PSNR and LPIPS values gradually decrease and increase respectively, as can be seen in Figure 6. The predicted trajectory is still similar to the actual game in terms of content and image quality, but per-frame metrics are limited in their ability to capture this (see Appendix A.1 for samples of auto-regressively generated trajectories).