How the rotation of your image affects the model's training
In a nutshell, probably not. If you are using AI to perform simple image differentiation, then most likely the orientation of your image will not affect the training of your model. For example, training a block to identify breeds of dogs or types of documents will work just fine even if some of your data is sideways.
However, if the direction of your image is important to the prediction of the model, for example, trying to identify from photos of eyes whether the eye is on the left or right, this may affect the quality of your training. In this case, you may want to correct the EXIF data of your images to ensure they have the correct orientation, before importing them into Levity.