Cloud schedulers, autoscalers, and capacity planners require realistic workload traces, but public traces are scarce, privacy-sensitive, and often too static to stress-test Internet-scale systems. We demonstrate Hyper-TS, a trace generation tool that learns from production virtual-machine traces and emits controllable synthetic workloads. Hyper-TS separates workload simulation into two coupled stages: a non-homogeneous Poisson arrival model for job batches and a conditional diffusion generator for VM instance types and job lifecycles. A disentangled temporal decoder preserves long-term trends and periodic bursts, while hybrid survival-aware lifecycle bins retain precision for short jobs and robustness for long-tailed, right-censored jobs. On the Microsoft Azure VM trace, Hyper-TS achieves 85.45% p05–p95 coverage and 0.3412 KL divergence for instance-type generation, and 16.88% 1-Best-Err with 0.1544 KL divergence for lifecycle modeling. The demonstration shows how users train the generator, control burst intensity and tail duration, visualize fidelity metrics, and export traces for scheduler testing. The open-source code is as follows: https://anonymous.4open.science/r/Hyper-TS-B1E5/