The JWST/MIRI mid-infrared deep field imaging is expected to reveal many thousands of high redshifted galaxies. To be ready for the morphological studies of these galaxies a simulation of the deep field is being prepared using the MIRISIM simulator. Since running the full astrophysical simulation of the sources would be very computer demanding, a new machine learning approach based on the use of a Variational AutoEncoder (VAE) is tested to generate examples of the early galaxies of different morphologies. The AE is also used to detect morphological anomalies/irregularities in the images of galactic core that may be related to an existence of a black hole binary system and a post merger product. Here I describe the method and show the preliminary results on dummy data.