Does no one care about these absurd content restriction policies?

Question

Does no one care about these absurd content restriction policies?

CodeKidz 25

Look at the prompt, even the simplest prompt can be filtered, and it shows that no rules have been violated.

This is obviously a very serious bug, which almost makes Dall-E completely unusable on Azure, and also affects the gpt model. I have reported this issue many times, but no one seems to care.

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': 'A basic stick figure that resembles the supplied sketch.'}, 'message': 'This request has been blocked by our content filters.', 'type': 'invalid_request_error'}}

AshokPeddakotla-MSFT 36,021 Reputation points Moderator

2024-09-16T04:30:47.1633333+00:00

CodeKidz Greetings!

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': 'A basic stick figure that resembles the supplied sketch.'}, 'message': 'This request has been blocked by our content filters.', 'type': 'invalid_request_error'}}

Are you seeing this issue with DALLE-3 or DALLE-2 model? Are you seeing this issue in Azure OpenAI Studio?

I just tried with the same prompt and able to get the image with DALLE-3.

Please note that, the image generation APIs come with a content moderation filter. If Azure OpenAI recognizes your prompt as harmful content, it doesn't return a generated image. For more information, see Content filtering and adjust the Severity.

Do let me know if that helps or have any further queries.
AshokPeddakotla-MSFT 36,021 Reputation points Moderator

2024-09-19T10:41:44.4533333+00:00

CodeKidz Just checking to see if you had a chance to review my response.

Do let me know if that helps or have any other queries.
CodeKidz 25 Reputation points

2024-09-19T11:38:03.42+00:00

The prompt in the message is extended by GPT automatically, I can't remember the original prompt, may be "A stick figure.', much shorter than the extended one.

You can use the data returned by the API to view the reversed prompt, or find some ways to prevent it from extending the prompt.

I can't find the original prompt now, so I can't reproduce the problem for you, but it occurs very frequently, so next time I encounter it, I will paste the original prompt here.

Beside, I use the v2 version of content filter, the default one is even worse.
CodeKidz 25 Reputation points

2024-09-20T13:39:42.7533333+00:00

@AshokPeddakotla-MSFT I've pasted some errors I met recently, the first line is the original prompt, the second is the error message, every rule was False and safe, but still blocked.

Definitely bugs!

a simple, brief, cartoonish image suit for a badge of Create your first story book and let your imagination soar!

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': "A cartoonish illustration designed for a badge. It should depict the concept of 'Creating your first story book'. Possibly include a child of ambiguous gender and of Hispanic descent, sitting down and writing in a large, open book. Above the child, a thought bubble with various classic story elements such as a castle, a dragon, a spaceship, etc. to symbolize the soaring imagination. The text 'Create Your First Story Book and Let Your Imagination Soar!' could float around in a decorative ribbon or banner."}, 'message': 'This request has been blocked by our content filters.', 'type': 'invalid_request_error'}}

a simple, brief, cartoonish image suit for a badge of Switch your teacher for the first time. Variety is the spice of life!

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': "Create a cartoonish image suitable for a badge, representing the theme 'First Time Switching Your Teacher'. It should depict a cheerful student with a backpack, opening a door signifying the change of teachers. Diversity is highlighted by having a mixed-race male student and a Caucasian female teacher welcoming him in the new classroom."}, 'message': 'This request has been blocked by our content filters.', 'type': 'invalid_request_error'}}

a simple, brief, cartoonish image suit for a badge of Complete twenty lessons and master the art of learning!

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'profanity': {'detected': False, 'filtered': False}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': "A cartoonish illustration suitable for a badge. The picture consists of an open book with 20 page markers along with a graduation cap resting on top, signifying the completion of 20 lessons and mastery in learning. The backdrop of the illustration is filled with light rays emanating from the book, symbolizing the enlightening power of education. Additionally, include an overlay text saying, 'Complete twenty lessons and master the art of learning!'. Ensure the image is simple and brief, ideal for a badge."}, 'message': 'Your request was rejected as a result of our safety system. The revised prompt may contain text that is not allowed by our safety system.', 'type': 'invalid_request_error'}}
AshokPeddakotla-MSFT 36,021 Reputation points Moderator

2024-09-20T14:20:02.71+00:00

CodeKidz I tried with every prompt you shared and not seeing any issues.

P.S. I did not configure any content filters.

I would suggest you try adjusting the Severity filter once and let me know if that solves your issue.

What do you mean by The prompt in the message is extended by GPT automatically?

Are you using some SDK or code to develop or improve the images?
CodeKidz 25 Reputation points

2024-09-21T02:45:36.06+00:00

There is no way to change the content filter level, and even in the playground, I'll meet these errors after try one prompt for 2-3 times, e.g. a simple, brief, cartoonish image suit for a badge of Create your first story book and let your imagination soar!

"The prompt in the message is extended by GPT automatically?"

I mean the original prompt is extended to the reversed prompt, as you can see in the response of API.
CodeKidz 25 Reputation points

2024-09-21T03:14:30.23+00:00

I found that the playground will not automatically extend the prompt, but you can't turn it down in the API request.

You can try with the reveresed_prompt in the json below. There is a high probability of encountering filtering. and every rule is safe in the response.

I have changed the content filter level to low for every item, but it didn't work. about 1/10 of my request will be rejected, that's quite a lot of rejections. However, my prompt is only to create some harmless badge images for some rewards, just like the one below.

I've recorded a video here: https://vimeo.com/1011518458?share=copy#t=0

a simple, brief, cartoonish image suit for a badge of Reset a completed lesson. Time traveler!

Error code: 400 - {'error': {'code': 'content_policy_violation', 'inner_error': {'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'hate': {'filtered': False, 'severity': 'safe'}, 'profanity': {'detected': False, 'filtered': False}, 'self_harm': {'filtered': False, 'severity': 'safe'}, 'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': False, 'severity': 'safe'}}, 'revised_prompt': 'Create a cartoonish image suitable for a badge symbolizing the resetting of a completed lesson. Incorporate elements of time travel into the design, such as a clock, hourglass or a swirling portal.'}, 'message': 'Your request was rejected as a result of our safety system. The revised prompt may contain text that is not allowed by our safety system.', 'type': 'invalid_request_error'}}
AshokPeddakotla-MSFT 36,021 Reputation points Moderator

2024-09-25T05:37:28.3233333+00:00

CodeKidz Got it.

Regarding revised_prompt, Prompt transformation is applied by default to all Azure OpenAI DALL-E 3 requests. After prompt transformation is applied to the original prompt, content filtering is applied as a secondary step before image generation.

Since you are getting the issue with low content filters, please try to provide the feedback directly on the portal. Please switch to new look to get that option.

Also, For a deeper investigation and immediate assistance on this issue, please file a support request @ https://aka.ms/azsupt?

Do let me know if you need any further assistance.