How can I ensure that my custom field's *to_python()* method is only called when the data in the field has been loaded from the DB?

I'm trying to use a Custom Field to handle the Base64 Encoding/Decoding of a single model property. Everything appeared to be working correctly until I instantiated a new instance of the model and set this property with its plaintext value...at that point, Django tried to decode the field but failed because it was plaintext.

The allure of the Custom Field implementation was that I thought I could handle 100% of the encoding/decoding logic there, so that no other part of my code ever needed to know about it. What am I doing wrong?

(NOTE: This is just an example to illustrate my problem, I don't need advice on how I should or should not be using Base64 Encoding)

def encode(value):
    return base64.b64encode(value)

def decode(value):
    return base64.b64decode(value)


class EncodedField(models.CharField):
    __metaclass__ = models.SubfieldBase

    def __init__(self, max_length, *args, **kwargs):
        super(EncodedField, self).__init__(*args, **kwargs)

    def get_prep_value(self, value):
        return encode(value)

    def to_python(self, value):
        return decode(value)

class Person(models.Model):
    internal_id = EncodedField(max_length=32)

...and it breaks when I do this in the interactive shell. Why is it calling to_python() here?

>>> from myapp.models import *
>>> Person(internal_id="foo")
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/usr/local/lib/python2.6/dist-packages/django/db/models/base.py", line 330, in __init__
    setattr(self, field.attname, val)
  File "/usr/local/lib/python2.6/dist-packages/django/db/models/fields/subclassing.py", line 98, in __set__
    obj.__dict__[self.field.name] = self.field.to_python(value)
  File "../myapp/models.py", line 87, in to_python
    return decode(value)
  File "../myapp/models.py", line 74, in decode
    return base64.b64decode(value)
  File "/usr/lib/python2.6/base64.py", line 76, in b64decode
    raise TypeError(msg)
TypeError: Incorrect padding

I had expected I would be able to do something like this...

>>> from myapp.models import *
>>> obj = Person(internal_id="foo")
>>> obj.internal_id
'foo'
>>> obj.save()
>>> newObj = Person.objects.get(internal_id="foo")
>>> newObj.internal_id
'foo'
>>> newObj.internal_id = "bar"
>>> newObj.internal_id
'bar'
>>> newObj.save()

...what am I doing wrong?