New Imaging capabilities VERY curtailled

[jsbinette]

After the announcement of new capabilities in terms of image generation, I gave gpt and old prompt I picked in my history.
It refused to do it on “content policy”. Yet, by it’s own anmisson, there was nothing wrong with the prompt. Even after GPT itself reformulated it, it couldn’t get it done.

Now that even Facebook is relaxing it’s policies; it’s OpenAI’s turn to get all woke?

Here is the old prompt and OLD image:

“Draw: A luxurious, exotic Bill Bensley-inspired interior design for a high-rise penthouse living room with floor-to-ceiling windows and a breathtaking city view. The space features rich, layered textures, vibrant patterns, and a mix of opulent NOT ornemental materials. A lush jungle-inspired wallpaper or mural on one wall, a dramatic statement chandelier, and an eclectic mix of sculptural seating. Modern wooden side tables in oak, plush velvet cushions, and a large patterned rug bring warmth and depth. Unique, oversized indoor plants and handcrafted decorative elements like golden futuristic statues, elaborate vases, and a statement coffee table anchor the space. The lighting is moody and atmospheric, emphasizing a blend of old-world grandeur and playful, maximalist whimsy.”

[jsbinette]

After the announcement of new capabilities in terms of image generation, I gave gpt and old prompt I picked in my history.
It refused to do it on “content policy”. Yet, by it’s own anmisson, there was nothing wrong with the prompt. Even after GPT itself reformulated it, it couldn’t get it done.

Now that even Facebook is relaxing it’s policies; it’s OpenAI’s turn to get all woke?

Here is the old prompt and OLD image:

“Draw: A luxurious, exotic Bill Bensley-inspired interior design for a high-rise penthouse living room with floor-to-ceiling windows and a breathtaking city view. The space features rich, layered textures, vibrant patterns, and a mix of opulent NOT ornemental materials. A lush jungle-inspired wallpaper or mural on one wall, a dramatic statement chandelier, and an eclectic mix of sculptural seating. Modern wooden side tables in oak, plush velvet cushions, and a large patterned rug bring warmth and depth. Unique, oversized indoor plants and handcrafted decorative elements like golden futuristic statues, elaborate vases, and a statement coffee table anchor the space. The lighting is moody and atmospheric, emphasizing a blend of old-world grandeur and playful, maximalist whimsy.”

[jsbinette]

This is what GTP proposed for my post:
Title: Why is OpenAI’s “Improved” Image Model Flagging Tasteful Interior Design Prompts?
Just a heads-up to the team: your latest image model was announced as more capable and nuanced — but it still flagged a fully appropriate, tasteful prompt for generating an artistic, Bill Bensley–inspired penthouse interior.

Here’s the original prompt:

“A luxurious, exotic Bill Bensley-inspired interior design for a high-rise penthouse living room with floor-to-ceiling windows and a breathtaking city view. The space features rich, layered textures, vibrant patterns, and a mix of opulent NOT ornamental materials. A lush jungle-inspired wallpaper or mural on one wall, a dramatic statement chandelier, and an eclectic mix of sculptural seating. Modern wooden side tables in oak, plush velvet cushions, and a large patterned rug bring warmth and depth. Unique, oversized indoor plants and handcrafted decorative elements like golden futuristic elements, elaborate vases, and a statement coffee table anchor the space. The lighting is moody and atmospheric, emphasizing a blend of old-world grandeur and playful, maximalist whimsy.”

Nothing remotely inappropriate here — just high-concept, detailed design with imaginative flair. Yet I received a content policy violation with no clear explanation why. Even after toning down the language and removing anything remotely edgy, it still got flagged.

This isn’t just annoying — it directly contradicts the promise of “better prompts, better understanding.” If the system can’t handle rich, poetic, or layered design language without throwing a flag, then it’s not really smarter — just more cautious.

Meanwhile, Meta is relaxing its filters and enabling creativity, and here I am getting blocked for suggesting velvet cushions and jungle murals.

Can someone from the team please look into:

• What specifically triggers the flag in prompts like this?

• Whether creative design-related language is getting caught in the net?

• How this aligns with the new image model capabilities that were just announced?

Because if “golden futuristic elements” and “moody lighting” are too much… we might need to redefine “advanced.”

[jsbinette]

And it refuses to make this too:
“Draw image: Bill Bensely inspired. Ultra modern. An extra small ultra luxury office with black and white floor. The office has blue walls with white moldings and a floor to ceiling window (curtain wall style with anodized frame) at the end. The office features a desk facing the wall with an office chair and a chaise angled towards the window. The ceiling is black and white and the floor is matte black and white.”

[_j]

@jsbinette

It refused to do it on “content policy”.

I suspect that despite gpt-4o being the one to create images, there is an inspection layer of the input being asked for. In this case, you are asking the AI to replicate the work of an existing creator, which may be exactly the kind of production that OpenAI would have a safety (public relations) system be blocking.

The only thing you were missing was an explanation of that justification, such as showing here following your pattern (although the AI may be inferring and elaborating on simply the words “content policy” received back):

---

---

I replicated the refusal. Substitute words for the named individual reveals success. Hover for the words.

ChatGPT with it’s dalle tool usage guidelines it has to read, may already have been replacing the name for you.

[jsbinette]

You are very right. I removed the name and it produced almost the same image as you got.

[ixahmed]

I am having same issue since yesterday. The same prompt to create a stained glass image was sharper in detail and quality with an image size of 1024 by 1972. Now the same prompt created a lower quality image of 1024 by 1536 image resolution. We do expect consistency when we are paying for ChatGPT plus.