Yes, Paillier encryption is secure from known plaintext attack (for single-character message, and any other supported message size). With high likelihood, three ciphertexts $c_1$, $c_2$ and $c_3$ for the same plaintext will be different.
When using public modulus $n$, each Paillier encryption draws a uniformly random number $r$ in range $[0,n)$ (some descriptions say $[0,n^2)$ but it turns out only $r\bmod n$ has an influence on the cryptogram; and in some descriptions it's added $r$ is coprime with $n$, but that's overwhelmingly likely for secure choice of $n$, thus can be omitted). Different $r$ will lead to different cryptograms $c=g^m\cdot r^n\bmod n^2$.
For proper implementation, by the birthday problem 101, the probability is about $k^2/2n$ that among $k$ ciphertexts for the same plaintext, any two are equal. For security against factoring, $n$ must have some thousand bits, thus said probability is entirely negligible for any feasible $k$ and secure choice of $n$. It is much more likely that whatever generates $r$ gets defective, and hickups, leading to identical ciphertexts.
Note: Pailler encryption is homomorphic only for messages that are straight integers $m$. Padding with zeroes on the left can't change $m$ if the homomorphic property must remain.