We adapt a CompVis latent diffusion model for 4× ERA5 weather downscaling, fine-tuning VQVAE + LoRA UNet on multichannel WeatherBench zarr data. A custom power spectral density (PSD) loss and alternative latent initialization improve fine-scale reconstruction over standard MSE training.
Tech Stack
PythonUncategorized
Why diffusion-weather-downscaling?
Limitations
6/17/2026
Last Update
0
Forks
0
Issues
Unknown
License
Financial Leak Detected
Stop the "SaaS Tax"
Your team could be burning cash. Switching to diffusion-weather-downscaling instantly boosts your runway.
