from Hacker News

Mirrors: The Blind Spot of Image and Video Generation Models

by yamrzou on 3/31/25, 4:30 PM with 3 comments

• by 8-prime on 4/3/25, 9:02 AM

  The generated images have an interesting surrealist feeling to them.

  This also makes me think that, for now atleast, including mirrors in photos/videos would be pretty potent to help prove that something is not AI generated.

• by smusamashah on 4/3/25, 10:10 AM

  These models are getting better and better. Mirrors will be solved too, guess just need more data. Water reflections for landscape images are already fine.

• by Mistletoe on 4/3/25, 8:08 AM

  The first Veed video genuinely terrified me.