Client
Region
Tech stack
We challenged bias in generative AI in partnership with Closeup (a Unilever brand) to build a custom AI models from scratch that was trained on diverse queer narratives across media, literature, and art to authentically reflect LGBTQ+ love and identity. The microsite hotsed a limited-time playful, personalized quiz that outputted unique AI-generated image and poem that celebrates love in all its complexity.
Making AI models inclusive
In collaboration with Socialcontext, we developed a bespoke large language model and stable diffusion image generator trained exclusively on queer media, literature, and art. Users answered a short set of prompts, and our system transformed their responses into a personalized poem and image. Users could download the image separately, or with the poem. This inclusive expression of intimacy challenged bias in AI and affirmed the closeness at the core of every connection.
Problem
Custom AI model trained on queer representation
We developed the image and text models from scratch using a curated dataset of LGBTQ+ stories, artworks, and cultural references. This required close coordination between data ethics and engineering to balance authenticity with clean, brand-safe results.
Personalized love artifact
The experience used two custom generative AI models — one for imagery and one for text — to turn each quiz result into a unique “love artifact.” The image model produced an authentic, bias-aware portrait, while the text model crafted a short poem that echoed the user’s relationship dynamic via quiz inputs.
Built-in safeguards
Because the experience touched sensitive areas of identity and self-expression, we implemented robust safety systems. Through content filters, moderation layers, and output validation tools, we ensured respectful, bias-conscious results.
