I have standard-definition video that needs to be converted on-the-fly to HDR 1080p. My priority is true real-time performance, not batch rendering, so every millisecond counts. I’m open to an off-the-shelf model, a custom-trained network, or a hybrid approach—as long as the end result is latency-free HDR upscaling that preserves detail, reduces artifacts, and respects accurate color space conversion. Here’s what I need delivered: • A fully functional pipeline (CLI or minimal GUI) that accepts an SD video feed and outputs HDR 1080p in real time. • GPU acceleration leveraged through CUDA, TensorRT, DirectML, or a comparable framework. • Controls for key parameters—noise reduction strength, sharpening, tone mapping—so I can fine-tune output quality. • Clear build instructions and source plus brief documentation that explains model choice, performance benchmarks, and any training data references. I’ll test the solution with several live and file-based SD sources; the job is complete once the pipeline consistently holds real-time speed while maintaining HDR fidelity.