We will present the recent progress of generative AI for the design of photonic devices, including variable autoencoders and diffusion models, and latent space optimization. We also review our non-traditional way of implementing optical neural networks using data re-uploading technique originally proposed for quantum computing.