AES is a block cipher that operates on 128 bit blocks and for any messages (plaintexts) of other size than 128 bit one uses AES in some mode of operation, e.g., CBC considering the message as a sequence of 128 bit blocks (plus padding if required) or modes like CTR to turn AES into a stream cipher.
Anyways, the ciphertext which is output by AES when used with a particular mode of operation is a bitstring (sequence of bytes) of some size (size of the input message plus padding if necessary) and has as it is nothing to do with any character in any encoding.
How you display/store/transmit the ciphertext (the bitstring/sequence of bytes) and thus the set of possible characters that are used therefore depends on the character encoding that is used, e.g., utf 8.
For any given message (bitstring), the ciphertext produced by AES can be any possible bitstring, i.e., all bit patterns of the size of the ciphertext are possible. For instance, let us for simplicity assume the message to be 128 bit and let it be $0^{128}$, then the ciphertext can be any element of $\{0,1\}^{128}$. Consequently, when using a character encoding for the ciphertext, any character (bit pattern) supported by the chosen character encoding may appear somewhere in an AES ciphertext.
In you case, the ciphertext is BASE64 encoded and thus you observe exactly this set of characters when displaying the ciphertext. This particular encoding is a usual choice, since you do not run into encoding issues when dealing with on/transmitting ciphertexts between different platforms/systems.