Context-aware nudity filtering needed - Nordic and Indigenous examples

ChatGPT
,
1

Dear OpenAI,

I would like to provide feedback on how image generation models distinguish between nudity and sexual content.

At present, the model appears to rely too heavily on visible skin or exposed nipples as indicators of sexual content. This approach does not adequately account for cultural context, intent, or the overall presentation of an image. As a result, neutral and culturally accurate depictions may be unnecessarily restricted or altered.

I believe the model should distinguish nudity from sexualization.

Sexual content is defined primarily by context, intent, composition, and presentation rather than by the amount of exposed skin. Factors such as seductive posing, emphasis on sexual body parts, erotic lighting, explicit sexual acts, or clear sexual intent are much stronger indicators than simple anatomical visibility.

By contrast, many situations involve nudity that is clearly non-sexual, including:

- Finnish and Swedish sauna culture

- Documentary photography

- Indigenous cultural traditions and ceremonies

- Practical nudity related to climate or everyday life

- Medical and educational contexts

- Breastfeeding

- Body-positive portraiture intended to represent ordinary people realistically

For example, in Finland, spending time shirtless at a summer cottage, on a lakeside dock, in a sauna, or in a changing room is a normal part of everyday life. These situations are generally understood as non-sexual. Likewise, many Indigenous cultures traditionally include bare torsos in ceremonial, cultural, or practical contexts. Automatically treating such images as sexual can unintentionally distort authentic cultural representation.

The same principle applies more broadly to anatomy. The visibility of a body part alone should not determine whether an image is classified as sexual. Context should remain the primary consideration.

I would encourage continued development of moderation systems that evaluate the whole image instead of relying primarily on exposed skin. Relevant signals include:

- the subject’s activity

- body language

- facial expression

- camera framing

- lighting

- artistic or documentary intent

- overall context

Likewise, indicators of sexualization may include:

- seductive eye contact directed at the viewer

- erotic posing

- deliberate emphasis on breasts, buttocks, or genitals

- oiled or stylized presentation intended to sexualize the body

- explicit sexual behaviour or implication

A context-aware approach would improve both cultural accuracy and fairness while maintaining appropriate safeguards against genuinely explicit sexual content.

Thank you for considering this feedback.

Kind regards,

A user from Finland

2

Welcome to the community @juntusmatti,

I can :100: get behind the core of this feedback. I was born in Finland and currently live in another Nordic country, so the sauna example is especially familiar to me.

Nudity itself should not automatically be treated as sexual. Sauna culture is an ordinary part of life in Finland, breastfeeding a child is not sexual and nudity has been part of art throughout history.

At the same time, I think this may be undoubtedly difficult to implement safely. Not everyone views or uses images containing nudity in a neutral way and image models can be misused, for example, to create non-consensual deepfakes or abusive imagery involving minors. Cultural norms and expectations also differ considerably. Developers and companies such as OpenAI therefore have to maintain a constant balance between understanding context and providing safeguards that work responsibly across many cultures and use cases.

That said, I hope this will be read and considered as meaningful feedback for future image models. Improving their ability to distinguish ordinary, cultural, medical, documentary and artistic nudity from sexualized content would be genuinely valuable.

Small note: I am a another user here and not an OpenAI employee.

3

Welcome to the community, @juntusmatti, and thanks for laying this out so clearly. Your request is for image models to distinguish non-sexual nudity in cultural, medical, documentary, artistic, and everyday contexts from genuinely sexualized content.

@LarisaHaster also raised an important point: not everyone views or uses nude imagery neutrally, so any more context-aware approach still needs strong safeguards against misuse.

We’re sending this feedback to the team for logging and consideration.

-Mark G.