I'm going to expand on Paŭlo's comment whilst I'm here:
The complete file from the outside is just a stream of bytes. Your encryption program doesn't (shouldn't) care if this originally was a word file with text and images, a plain text file, a video file or something else. You just encrypt the bytes as they come.
What you're not understanding, I think is the concept of a block, or a bit. In classical cryptography, it was common to use data types such as letters and numbers - this is what humans store and process, after all. Statements such as "take A, shift it three letters down the alphabet to D" are things humans can understand.
Computers can understand these too - but they also represent a lot of other types and formats of information - much more than humans. They do this in binary - strings of zeros and ones. Each digit is called a bit, 8 bits is called a byte.
So now we get onto the concept of a block. A block is just a collection of bits, or bytes. A 128-bit block, which AES uses, is 128 bits, or 16 bytes.
Now, just to confuse you, there are several types of AES encryption - 128-bit, 192-bit and 256-bit. These refer to the size of the key - not the input block. That block always, always stays at 128-bit, at least for AES (Rijndael, the algorithm on which it is based, supports more block sizes).
So, when Paŭlo says "take the bytes as they come" you literally read the file block by block - 16 bytes at a time. However, do please keep reading...
You can just encrypt these blocks byte at a time. AES can and will do that and the mode is called electronic code book (ECB). However, now I'm going to translate Hendrik's (very true, +1) statement:
AES is just a standard block cipher primitive. The only thing this standard tells you, is how to transform a single 128 bit block using a key, into another 128 bit block. It doesn't tell you where the key is coming from, or how you are going to process any data larger than 128 bits.
See here's the thing. In our ECB mode, we handle files larger than 128 bits by simply dividing the file up into 128-bit chunks. It turns out this can be quite an insecure way to work, as patterns might emerge in the ciphertext which relate to the file structure.
As such, there are different ways to use a block cipher called Modes of Operation. Hence why Hendrik referred to AES as a primitive - because it only forms part of a larger scheme. There are many modes of operation and not all of them will suit your use cases - but needless to say you probably want one that is not ECB.
These new modes of operation still take files in terms of bytes and bits - sometimes in 128-bit blocks, sometimes bit at a time, depending on the mode in question. The concept of using the computer's native formats (bits and bytes) rather than text, remains the same and thus any file type can be handled.
As Hendrik has also alluded to - the problem of generating a key is also an interesting case. I've seen software developers simply convert a string (in binary terms) straight into a key. This is a really poor way to go about generating a key, since the number of possible keys is much, much more limited than the AES keyspace. I won't say any more here as this is about files - but needless to say it is definitely something to look into.