MyTimeMachine: Personalized Facial Age Transformation

The "Dreambooth" results seem much better than the authors'.

I think the claims here are kinda weak, not gonna lie. It's unclear to me that this means of adding case-by-case additional conditioning data to improve results is particularly novel or effective compared to the popular StyleGAN-based attribute walk the overall approach is based on (Though as a side note it's definitely fun to see this kind of thing again after it's fallen out of favor in recent years with all the attention large foundation models have been getting in the generative space), and as a machine learning result, as far as I can tell this is kind of just saying "if you have additional highly-relevant data you can slightly overfit your model to cover a specific case better"

Is there something I'm missing here?

For my case, I've seen some guys/girls who used to be very pretty/cute during my younger years (e.g. primary school) but later on become meh. And some who looked unremarkable at all suddenly "Is that really them???"

I wonder if there's an AI for that.

Even more useful: AI that gives you step by step from meh to wow. But I guess that's called make up class.