Accelerate scientific diffusion models without retraining
Generate 3D motion from audio with continual‑learning demo