Noise gone. Structure stays.
Our own ML model was trained on Seestar S50/S30 stacks with their real, measured noise. It works on the raw linear data — before the stretch, where noise can still be separated cleanly — and keeps stars and the faintest filaments intact.
- Three modes: Light, Medium, Strong
- Runs on the linear stack, not on the stretched image
- GPU-accelerated via DirectML, CPU fallback
Without
ML Denoise



