I don’t know since when, but something is seriously wrong with safety filters now. As reproducible in Google AI Studio, a simple and completely safe prompt fails miserably because of safety filters:
With the API, I get “finishReason: IMAGE_OTHER”.
As you can see, there’s absolutely nothing harmful about that prompt.
Yeah I had a similar problem, that appeared for no reason why I asked it to remake a meme of that little girl smiling in front of a burning house and all I asked was that it added a toilet paper roll into the girl’s hand, since I was using it as the artwork for a blog post I was writing on whether there is a right way to orient toilet paper (there is by the way, you can check the original patents if you do not believe me). And then Nana Banana was like “oh no sorry this is too violent I can make an image of a house on fire”. Like…
