Hi everyone,
I’ve recently been testing DALL-E 3’s capabilities for generating cityscapes and landscapes, but I’ve encountered some inconsistent and seemingly arbitrary restrictions. While some images generate just fine, others—particularly those involving certain architectural styles or geographical themes—are blocked.
What I Tested:
Kenyan Highlands – Successfully generated.
Sana’a inspired City – Blocked.
Western City – Successfully generated.
Eastern City – Successfully generated.
Middle Eastern City – Blocked.
Desert City – Blocked.
Pyramids in a Desert – Blocked.
Forest City – Blocked.
Dog – Successfully generated.
Western City – (Second attempt) Blocked.
Generic City – Successfully generated.
Observations:
- The restrictions appear location-based rather than related to realism or quality.
Major historical or cultural landmarks seem to be restricted (e.g., pyramids, Westminster-style buildings). - Even generic descriptions like “desert city” were blocked, despite being geographically neutral.
- This filtering is not consistent—some cityscapes are allowed while others are not, even if they follow the same prompt structure.
- The system does not provide feedback on why an image is blocked, making it difficult to adjust prompts accordingly.
Concerns & Suggestions:
Clarify content policies– Are entire regions restricted? If so, why?
- Improve prompt flexibility – It doesn’t make sense that “Eastern city” works but “Middle Eastern city” does not.
- Give better feedback when a request is blocked– Instead of a generic “error,” explain whether it’s due to content policy, technical issues, or something else.
- Re-evaluate architectural and landscape restrictions – Cities and landscapes are neutral subjects, and blocking them without explanation limits creative potential.
Has anyone else experienced similar issues? Would love to hear if there’s a workaround or if OpenAI can provide some insight into these restrictions.
Thanks!