How to generate synthetic feature instead of synthetic image?

[Chandan-IITI]

Is it possible to pass some extracted features from the pre-trained ResNet model to a diffusion model for training and further generate synthetic features instead of images like GAN or VAE?