High-quality 3D medical imaging data serves as the foundation of modern radiology AI, but long-standing "data bottlenecks" in training healthcare AI models persist due to constraints on data privacy, annotation costs, and sample sizes. To address this challenge, NVIDIA has recently launched the open-source framework NV-Generate-CTMR, which supports synthetic generation of high-resolution 3D CT and MRI data, enabling researchers to build more efficient medical imaging AI systems. Developed based on NVIDIA's previously released Medical AI for Synthetic Imaging (MAISI) architecture, the latest version, MAISI-v2, employs Latent Rectified Flow technology. This approach delivers approximately 33x faster inference compared to traditional diffusion models while enhancing generative quality. The framework can generate complete 3D volumetric medical images alongside corresponding anatomical structure segmentation results, making it applicable to tasks such as data augmentation, cross-modal synthesis, and tumor segmentation. Concurrently, NVIDIA announced the release of its brain MRI generation model, NV-Generate-MR-Brain. Trained on MR-RATE—the world's largest open-source multimodal MRI dataset—this model leverages over 83,000 patient records comprising approximately 700,000 MRI volumes, each paired with anonymized radiological reports and DICOM metadata. It supports generating multiple brain MRI sequences including T1, T2, FLAIR, and SWI, capable of producing full-brain images at resolutions up to 512 × 512 × 256 voxels. A distinguishing feature of NV-Generate-CTMR compared to conventional medical image generation methods is its support for variable voxel dimensions, diverse volume sizes, and whole-body range synthesis. This adaptability accommodates various clinical scanning protocols without requiring separate model retraining for different organs. NVIDIA positions this capability under the designation of a "medical imaging foundation model." The project is now fully open-sourced, providing pre-trained weights, training configurations, and inference code that allow researchers to deploy directly on local GPUs or perform fine-tuning. According to NVIDIA, synthesized medical imagery will play an increasingly vital role in areas such as privacy-preserving data sharing, simulation of rare pathologies, and enhancement of generalizability across medical AI applications.