I've been trying to get this embedding right forever. I'm pleased with the results at this point. This was made from 44 images that I took and does a good job of capturing the likeness. It does not produce outputs of a real person, rather features.44 images, 12 vectors per token, learning rate of .005, 10,000k steps, to be quite honest I could probably go with a lower learning rate as many produced good likenesses and quality. I hope you enjoy please review and share your images.

Files with SHA256: 81658c15f77b4be37c8a323a7ff1a144ee9c176e83800015ed48887d69e1ac1a