EO-WM

Official checkpoints for EO-WM: Toward Physically Informed World Modeling for Probabilistic Earth Observation Forecasting (NeurIPS 2026).

EO-WM is a weather-conditioned latent diffusion model for multispectral Earth observation forecasting. It uses 10 observed Sentinel-2 RGBN frames and meteorological forcing to predict 20 future frames at 128 × 128 resolution.

This repository provides the EO-WM master weights and the EarthNet2021-finetuned EO-VAE with its matching configuration.

GitHub — code, setup, and usage: https://github.com/hku-sail/EO-WM
Paper: https://arxiv.org/abs/2606.27277

eo-wm/master.safetensors
eo-vae/eo-vae.ckpt
eo-vae/model_config.yaml

Please follow the GitHub README for data preparation, training, and inference instructions.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for ll-13/eo-wm_weight

EO-WM: A Physically Informed World Model for Probabilistic Earth Observation Forecasting

Paper • 2606.27277 • Published Jun 25 • 3