What would be the best way to use a pre-trained autoencoder as an image input embedder?

[Kurtoid]

I already have the structure and weights for an image autoencoder. Should I do this as a InputFilter or a Input Embedder? How should I go about doing this?