SDG-SynHuman: Large-Scale Synthetic Video of Digital Humans

NVIDIA created a large-scale synthetic video dataset containing 236,937 clips totaling approximately 5,841 hours. The dataset features digital humans rendered in diverse indoor and outdoor 3D environments, with each sample being a temporally coherent 60-120 second video clip at 1080p and 30 fps. It was last updated on May 29, 2026.

Use Cases

• Training world foundation models based on synthetic human movement and interaction data
• Post-training model refinement based on temporally coherent video sequences
• Physical AI research based on diverse 3D environment renderings
• Computer vision model development based on high-resolution synthetic human video

Strengths

• 236,937 video clips provide a substantial volume of training data
• Approximately 5,841 hours of total video content
• High-resolution 1080p video at 30 frames per second
• Temporally coherent clips of 60-120 seconds each

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Data may reflect bias inherent to the synthetic generation methods and source environments

Provenance

Source

NVIDIA

Collection Method

Synthetically rendered 3D scenes

Freshness

Last updated 2026-05-29 17:47:49; freshness should be verified