Neural

Two neural nodes for the things traditional DSP can’t reach. Neural Stem Separator runs Demucs v4 to split a finished mix into four stems — drums, bass, other, vocals — useful when you want to sample one part of an existing recording and process it on its own. DDSP Resynth extracts pitch and loudness contours from a buffer and resynthesises it additively, opening up novel pitch-and-amplitude transformations that don’t exist in classical pitch shifting. Both nodes run entirely on your machine — no cloud, no API keys, no subscriptions, no data leaving the session. They’re slow compared to the rest of the categories (these are real neural networks, not real-time DSP), so K2K treats them as render-on-demand: hit the render button when you want a fresh result. Available only in builds compiled with K2K_USE_ONNX=ON. AI as a tool inside the graph, not the product.