I’ve been struggling with the intensely strict image safety guardrails as well.
I recently decided to try Flow to generate videos of fictional characters (fashion and lifestyle content, such as fashion selfies outfits like dresses) and every single time I try to generate a video of a character in a dress, the result is blocked because it’s apparently “harmful content”? I don’t understand how a character (a fictional woman) taking a selfie (simply described as smiling while taking the selfie) is somehow harmful?
I can only imagine the safety features have been overtuned, and will strike down any video generation of a woman in any form of fashion.