When using Stable Diffusion, you can see deformed faces when they are small (occupy small region of pixels). This is because of the quality of the VAE encoder and the decoder. Some details are lost when you encode and decode an image.

This simple workflow helps you investigate and study the effect of VAE in isolation. It will simply compress the input image and decompress it back. You can compare the input image and the output image to notice the artifacts.

This will let you know the theoretical maximum quality of the generated images.