Hello!
I am trying to use Google’s new ‘nano-banana’ model on Vertex AI with ALL safety filter settings set to off/block_none. But there seems to be an issue when its built-in image generation model thingy rejects it regardless with a prebuilt rejection message like “I’m just a language model and can’t help with that.” or “Sorry, I am unable to help you with that.” Even if the text generation before it agrees to generate the image, it gets cut off by one of those rejection messages.
Is this a bug, or was this intentional?
1 Like
Hey @TKR_Josh - are you still having this issue?
I am wondering if something like a copyrighted image might be what is triggering the safety filter, but you are getting a somewhat vague error message.
1 Like
I don’t think I’ve attempted to generate anything that would’ve been copyrighted by anything. I’m pretty sure it’s another safeguard inside the model that pretty much ignores the safety settings. I tried to get a guy from a cartoon to shoot a gun and I get one of those rejection messages.
It appears to be censored less on Google AI Studio, though AI Studio doesn’t have adjustable safety settings for generated images. Why does nano banana perform better on AI Studio and not Vertex AI?
Edit: Okay. The image safety settings are no longer on Vertex AI. What’s going on?
Flash 2.0 Image / Flash 2.5 Image (nano banana)
