Crazy Google Censorship

I tolerate it and adapted, but everything has its limits. How are you censoring? What did you censor from my prompt?

a full-body shot of Nina Broughton, a 26-year-old freelance photographer.

she has a confident and strong presence. her build is solid and athletic. her height is 178cm, weight 85kg. features are distinctly European.

she has blonde hair, styled in a casual, slightly tousled bob, and fair skin.

nina is dressed in her signature style: a simple black t-shirt, a high-quality black leather jacket, and well-fitting dark blue jeans. she stands in a firm, confident pose with her arms crossed over her chest.

her professional Canon EOS R5 camera is slung over her shoulder.

Include the name "Nina" somewhere visible within the image, in a stylish, modern font.

The setting is a modern Chicago street at night, with neon lights providing dramatic contrast lighting.

artistic style: highly detailed digital character art, cinematic, sharp focus, Octane render.
2 Likes

Check your safety settings first. They might be a bit strict if left untouched.
And I wonder if this censorship is different for others, as I got an image with that prompt!


Then I tried it through my ComicMaker App, which will generate multiple versions with some additional variations, but it uses Imagen 4 and I managed to get ten different versions. None of them were blocked. My app did notice that one image did not match the description, though. (Wasn’t in a Chicago street) so it redid that image. Another image also didn’t match the prompt exactly, so that is annoying.
In my experience, the censorship tends to be a little crazy sometimes. My ComicMaker app this has a “PG13” button that rephrases prompts in the hope it will pass through that way.
And strangely enough, I adjusted the text a little and included the word “nude” and still no censorship. The result was a woman in a swimsuit, but it was a result.
So, I asked my App to make a PG13 prompt from yours and this was the result:

A full-body depiction of Nina Broughton, a 26-year-old freelance photographer, exuding a confident and strong presence. She has a solid and athletic build, standing 178cm tall and weighing 85kg, with distinctly European features. Her blonde hair is styled in a casual, slightly tousled bob, complementing her fair skin. Nina is dressed in her signature style: a simple black t-shirt, a high-quality black leather jacket, and well-fitting dark blue jeans. She stands in a firm, confident pose with her arms crossed over her chest, her professional Canon EOS R5 camera slung over her shoulder. The name “Nina” is visibly incorporated into the image in a stylish, modern font. The setting is a modern Chicago street at night, with vibrant neon lights creating dramatic contrast lighting. Artistic style: highly detailed digital character art, cinematic, sharp focus, Octane render.

Does that work? It did for me.
So, I tried another prompt, asking for an image of a woman in body paint:

A captivating studio portrait of a young woman, fully body-painted in vibrant teal and orange hues, complemented by intricate spiral designs on her face and torso. Her blonde hair is styled in a braid. She is kneeling gracefully on a dark surface, her left hand gently touching a draped piece of purple fabric. The background features a striking abstract backdrop with large, swirling orange and purple spiral patterns, creating a dynamic and artistic composition with strong visual contrast.

That too worked fine and generated images that I can’t share here, but it was indeed just bodypaint covering a woman. But this time, I did get one blocked version as I got the error " Image Generation Failed (Attempt 2) The request was blocked by safety filters due to potentially sensitive content in the prompt." in my app.

So, the censorship isn’t just on the prompt, but also with the generated image. If the AI generates an image that it considers unsafe then it can still block it. It has generated it, but you don’t get to see it.

Hi @Setapca
I was also able to generate the image in the Gemini web application using the prompt you shared. Did you experience censorship when you tried it with the API?
Thank you