DALL·E (gpt-image-1) Prompt Rejected by Content Safety Filter – Safe Child & Puppy Interaction

Question

DALL·E (gpt-image-1) Prompt Rejected by Content Safety Filter – Safe Child & Puppy Interaction

Niket Kumar Singh 785

Hello Folks,

We are using the Azure OpenAI gpt-image-1 (DALL·E) model in Azure AI Studio for a legitimate and safe use case: generating illustrations for a children’s book involving a joyful scene of a child and her pet puppy on a beach.

Here is the exact input we're using:

GeneralDescription: "MAX jumps up, licking LUCY's face. She laughs, hugging him." LocationDescription: | MAX jumps up, licking LUCY's face. She laughs, hugging him. The scene is captured on a sun-drenched beach with soft, light golden sand. The ocean's edge is visible in the mid-ground, where gentle, clear turquoise waves lap the shore, glistening under the bright daylight. The sky is a vibrant, clear blue, casting a warm glow over the entire scene. The atmosphere is open, light, and joyful, highlighted by the brilliant sunlight. Characters: Name: "Lucy" Attributes: "Young Girl, 7, Fair Skinned" Position: "Seated on the sand, angled slightly towards the camera, with her arms wrapped around Max" Dress: "Light blue, short-sleeved t-shirt and vibrant red shorts, with her bare feet casually resting on the sand" Expression: "Head tilted back slightly, eyes closed in laughter, a wide, joyful smile on her face, with streaks of clear water or puppy drool visible on her cheek where Max is licking" Action: "Engaged in a warm, tight hug with Max, her body relaxed and joyful in the embrace" Name: "Max" Attributes: "Puppy" Position: "Perched on Lucy's lap and chest, with his front paws resting on her shoulders/collarbone area, his head positioned to lick her face" Dress: "Fluffy, light brown fur with lighter chest and paw markings, appearing well-groomed" Expression: "Mouth slightly open with his tongue extended, actively licking Lucy's cheek; eyes are bright and full of affection and playfulness" Action: "Actively licking Lucy's face, his body slightly squirming with excitement and affection, tail blurred from rapid wagging" CameraAngle: "Medium close-up, slightly low angle, capturing the intimate and joyous interaction between Lucy and Max. SHOTmdPBTTYzrkfm7Z1C"

Despite applying lowest threshold content filtering (LowThresholdTest1) and disabling prompt shields and blocklists, the system responds with: Uploaded image

What We Have Tried:

Deployed and applied custom filter (LowThresholdTest1) with:
- Lowest blocking on all 4 categories
- Prompt shields off
- No blocklists
Tried minor prompt rewording
Verified prompt contains no inappropriate, suggestive, or violent content
Prompt is consistent with wholesome children’s story themes

Can Microsoft please clarify if model-level image filtering for DALL·E 3 is non-configurable, even in Azure AI Studio?
Is there an official path to request a filter override for safe, creative use cases (e.g., children’s books)?
Does the https://aka.ms/oai/modified-filters link still function? It currently redirects to Bing or fails when not signed into the tenant.

Accepted answer

0 additional answers

Your answer

Answer 1

Hello Niket Kumar Singh,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having issues with DALL·E (gpt-image-1) Prompt Rejected by Content Safety Filter – Safe Child & Puppy Interaction.

To troubleshoot the content filtering issues, you will need to start by identifying the type of filtering that occurred. If the API response includes:

{ "error": { "code": "contentFilter", "message": "Your task failed as a result of our safety system." } }

then the prompt itself was blocked. But if it says:

{ "error": { "code": "contentFilter", "message": "Generated image was filtered as a result of our safety system." } }

then the image output was filtered. However, if no error appears and the image uploads successfully, the issue may have been resolved or bypassed by your current filter settings. In that case, verify whether an image was returned to confirm resolution.

Secondly, refine your prompt to avoid triggering filters. Even wholesome scenes can be flagged due to sensitive combinations of body parts, animals, or emotional expressions. For example, instead of using potentially problematic terms, rephrase your prompt like this:

GeneralDescription: "MAX bounces onto LUCY’s lap, nuzzling her cheek with joy. She laughs brightly, wrapping her arms around him."
Characters:
  Name: "Lucy"
  Attributes: "Young Girl, 7, Fair Skinned"
  Dress: "Light blue shirt and red shorts"
  Expression: "Laughing, eyes closed with joy"
  Action: "Hugging Max warmly"
  Position: "Sitting on beach sand, legs stretched"
  Name: "Max"
  Attributes: "Fluffy puppy, light brown fur"
  Expression: "Tongue out playfully, affectionate eyes"
  Action: "Pressing face affectionately against Lucy's cheek"
  Position: "Front paws on Lucy’s chest, wagging tail"
CameraAngle: "Medium close-up, slightly low angle, bright sunny beach scene"

Avoid terms like “licking,” “bare feet,” or “squirming”, as they may unintentionally trigger safety filters. If the issue persists, escalate it through the Azure OpenAI resource, via your Azure portal then navigate to Support + Troubleshooting to create new Support Request.

Lastly, note that the link https://aka.ms/oai/modified-filters is currently deprecated or restricted to internal use. If it redirects or fails to load, it’s not usable for public tenants, and no alternative has been provided as of mid of this year 2025.

I hope this is helpful! Do not hesitate to let me know if you have any other questions or clarifications.

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.

Share via

DALL·E (gpt-image-1) Prompt Rejected by Content Safety Filter – Safe Child & Puppy Interaction

0 additional answers

Your answer