domains / families / media / physical-ai-event-video-generation

physical-ai-event-video-generation

separate

Run the PAIDF Orchestration Event Video Generation DAG on Kubernetes - image-to-video anomaly generation, auto-labeling, and anomaly dataset generation. Select for requests about event video generation, anomaly video generation, image-to-video synthesis, Cosmos3 image2video, anomaly dataset creation, safety/surveillance SDG, or generating person-falling, person-climbing, person-running, fighting, smoking/vaping, fire/smoke, or shoplifting video clips from a seed image. Runs environment setup first when controller readiness is unknown. Not for person-crop clothing/attribute augmentation (that is image-attribute-augmentation-workflow) and not for video style transfer.

When to use

• **Controller pods** — check that the Airflow controller pods (not DAG task pods) are Running.
• **Airflow API** — reachable only if check 1 passes. First establish `AIRFLOW_URL` from the
• **Pools** — only if check 2 passes. Required pools with open slots: `k8s_gpu_1`,
• **Compute-cluster GPUs** — check the cluster (using the cluster connection established above):
• **Task pods that run local model inference regardless of service mode**:
• **Internally deployed endpoints** (`external_services: false`): one GPU per VLM/LLM replica,
• **Internal mode total = both sources combined**: the three task-pod GPUs plus the four
• **Stale failed pods** — before triggering, check for accumulated failed pods in the compute

When not to use

• person-crop clothing/attribute augmentation
• video style transfer
• video generation without auto-labeling

Skill page

Source