In fact, *every* steganographic method can be broken with currently available stegoanalitic methods(which are typically statistical methods). The only question is whether there is enough data, but practical amounts are sufficient (no astronomical figures like in crypto).
I think it depends how you do it.
Imagine you have your "original" image that you want to modify to include your key via steganography. If your original image satisfies certain characteristics, I am sure you couldn't tell if the post-processed image contains steganography or not.
Here is how I construct my "original" image:
* take a camera picture.
* add noise on it and save as *.png (lossless format)
The noise addition follows this algorithm: For each pixel take each original 8bit rgb value and replace the LSB (rightmost least significant bit) by a random 0/1.
And here's how I hide my 256 bit bitcoin private key (pk) inside this image:
* calculate y = "pk" XOR "sha256(myPassword)"
where XOR is a bit-wise operation of 256 bits.
y is a 256 bit sequence that looks absolutely random, also from statistical analysis point of view.
* I replace the first 256 "noise LSBs" of the image by "y".
Done!