Hacker News new | past | comments | ask | show | jobs | submit login

The man/woman color inversion one was the most impressive to me. On the rotations, I can rotate in my mind and see the other view… but I find it very hard to color invert mentally



That is amazing, here's the link for anyone interested (there's a lot of images on that page)

https://dangeng.github.io/visual_anagrams/static/videos/grid...


For me it's the reverse: the color inversions feel hardly more impressive than the morph animations that were all the rage in the 1990, because while I certainly understand how straight-forward color inversion is on the level of pixel data, I still can't "see" that simplicity. It hardly looks any different than an alpha blend with no relation at all.

The rotations on the other hand, wow! It is perfectly visible how the pixels don't change. You can physically rotate the screen and the image "changes". I could not think of a better illustration of how diffusion model images are not just echoes of preexisting images (they certainly are), but solutions to the problem of "find a set of pixels that will match the description of {prompt}". Or in this case, "that will match {A} when oriented this way and {B} when oriented that way".


I can see the woman when the man is shown if I look for it, but not vice versa, for whatever reason.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: