Pulse of Motion Benchmark
Measuring temporal realism in AI video generation. Does your model produce motion that matches real-world physics? The PoM benchmark evaluates this using PhyFPS — a metric that predicts frame rate directly from visual dynamics, without reading metadata.
100
Benchmark Prompts
4
Evaluation Metrics
24 FPS
Reference Frame Rate
Open
Submission Access
What We Measure
PhyFPS (Physical Frames Per Second) captures how closely AI-generated video motion matches the temporal dynamics of the real world. A model with low PhyFPS error produces videos where objects move at physically plausible speeds.
PhyFPS MAE
Mean Absolute Error between predicted and expected physical frame rate. Lower indicates the model produces more temporally realistic motion.
PhyFPS MAPE
Mean Absolute Percentage Error. Normalized version of MAE for cross-comparison across different frame rate ranges.
Intra-Video CV
Coefficient of Variation across sliding windows within each video. Lower means more temporally consistent motion throughout the video.
Text-Video Alignment
CLIP-based similarity between the input text prompt and the generated video content. Measures whether the generated video matches the intended description. Submissions below a threshold may have significant text-video misalignment.
Note: This is a supplementary metric, not a primary evaluation dimension. We set 0.16 as the threshold — submissions scoring below this value may lack meaningful text-video alignment and should be interpreted with caution.
How It Works
Four simple steps to benchmark your video generation model.
Download prompts
Get the 100 benchmark prompts from the Prompts page.
Generate videos
Run each prompt through your model to produce one video per prompt.
Upload submission
Pack all 100 videos into a .zip and upload through the Submit page.
Get ranked
Our pipeline evaluates PhyFPS and ranks your model on the leaderboard.