Update ReadMe

README.md

@@ -73,7 +73,7 @@ model = BACKBONE_REGISTRY.build(
 )
 ```
 
-The model supports the following raw inputs which you can specify in modalities: S2L2A, S2L1C, S1GRD, S1RTC, DEM, RGB. 
+The model supports the following raw inputs which you can specify in `modalities`: S2L2A, S2L1C, S1GRD, S1RTC, DEM, RGB. 
 If your data does not use all bands of a modality, you can specify a subset with `bands={'S2L2A': ['BLUE', 'GREEN', 'RED', 'NIR_NARROW', 'SWIR_1', 'SWIR_2']}`. 
 You can pass the inputs as in a dict to the model. If a tensor is directly passed, the model assumes it is the first defined modality. 
 TerraMind can also handle missing input modalities.
@@ -131,12 +131,13 @@ model = FULL_MODEL_REGISTRY.build(
     modalities=['S2L2A'],
     output_modalities=['S1GRD', 'LULC'],
     timesteps=10,  # Define diffusion steps
-    standardize=True,  # Automatically applies the standardization values to the input and output
+    standardize=True,  # Apply standardization
 )
 ```
-Like the backbone, pass multiple modalities as a dict or a single modality as a tensor to the model which returns generated images as a dict of tensors.
+Like the backbone, pass multiple modalities as a dict or a single modality as a tensor to the model which returns the generated `output_modalities` as a dict of tensors.
 Note: These generations are not reconstructions but "mental images" representing how the model imagines the modality.
 You can control generation details via the number of diffusion steps (`timesteps`) that you can pass to the constructor or the forward function.
+By passing `standardize=True`, the pre-training standardization values are automatically applied to the input and output.
 
 We provide an example notebook for generations at https://github.com/IBM/terramind.