Hello OpenAI team,
I’m currently working on a project that relies heavily on the DALL·E API for generating character-driven visuals with artistic flair – mainly fashion-forward, expressive portrait work. All of this is done within safe, non-explicit boundaries (no nudity, no suggestive clothing), in a style you’d typically see in editorial photography, artbooks, or video game concept art.
Recently, I’ve noticed a serious limitation that seems to disproportionately affect generations involving female characters.
Reproducible API Behavior:
I ran the following two prompts (same structure, same intent), and here are the outcomes via the DALL·E API:
- Prompt A: A close-up portrait of a woman, blushing, looking up, mouth slightly open, tongue out, drooling. Face only.
→ Blocked - Prompt B: A close-up portrait of a man, blushing, looking up, mouth slightly open, tongue out, drooling. Face only.
→ Accepted
Both prompts are identical in form and have no explicit content — just facial expressions. Yet only the female subject gets flagged.
The same happens with more abstract prompts involving crop tops, elegant dresses, or feminine body proportions — often blocked even when fully clothed — whereas the male equivalent typically passes. The inconsistency is interfering with artistic and creative workflows.
Why I’m posting this here:
I looked extensively for a DALL·E-specific category on the forums to submit this feedback directly to the image model team, but I couldn’t find one — which is why I’m posting under the API category instead. (If there is a better place, feel free to redirect me.)
You can find my initial, more community-oriented post (aimed at artistic fairness and feedback) titled:
“Gender asymmetry in DALL·E filters? Artistic test shows possible bias”
under the Community section of this forum.
Suggestion:
Could the filtering be adjusted to evaluate context and composition rather than keywords or subject gender? It might be helpful to have a manual review option for borderline prompts rather than an automatic block.
This isn’t a complaint about safety — I appreciate OpenAI’s commitment to responsible AI — but I hope this feedback helps improve balance and fairness for all creators.
Thanks for reading and for all your work.
— A concerned but respectful creator