Add inference result images to model card

README.md

@@ -309,6 +309,22 @@ fig.colorbar(im, ax=ax, shrink=0.7, label="dBZ")
 plt.savefig("refc_prediction.png", dpi=150, bbox_inches="tight")
 ```
 
+## Example Inference Results
+
+4-hour autoregressive forecast initialized from HRRR analysis on 2026-03-13 00Z, with GFS conditioning. Showing 2m temperature, 10m U-wind, composite reflectivity, and mean sea level pressure.
+
+**Initial conditions (+0h):**
+
+![StormCast CONUS +0h](images/stormcast_conus_2026-03-13_lead0h.png)
+
+**+2h forecast:**
+
+![StormCast CONUS +2h](images/stormcast_conus_2026-03-13_lead2h.png)
+
+**+4h forecast:**
+
+![StormCast CONUS +4h](images/stormcast_conus_2026-03-13_lead4h.png)
+
 ## Limitations
 
 - **Regression-only:** No diffusion model, so predictions are spatially smooth. The full StormCast pipeline requires the EDM diffusion stage for realistic fine-grained storm structure (convective cells, precipitation bands).