Microsoft is the technology partner behind the Community Library. They developed the platform – called Community Library Maker – that enables the amputee community to define, curate, and annotate the image dataset. Microsoft Research leads the technical process and ensures the final dataset is published correctly on an open-source platform.

The initiative is led by Ottobock, the world's leading manufacturer of prosthetics and mobility solutions. The library itself is built in collaboration with Ottobock’s global community of amputee ambassadors.

The Dear AI library is a curated image and video dataset built by and for the amputee community. The purpose is simple: to give AI systems accurate, authentic, and dignified visual references of people with limb loss and differences, so that when AI generates images of this community, it does so correctly.

It was built through a structured three-phase process on Microsoft's Community Library Maker platform.

In Phase 1, community project leads defined what good representation looks like – selecting reference images and identifying key themes like family life, work, sport, and daily independence.

In Phase 2, community members uploaded and annotated up to 400 images and videos per community, describing what is shown and why it represents them well.

In Phase 3, the community evaluated AI-generated images against their themes, rating their accuracy.

Two separate libraries are being created, one for lower limb amputees and one for upper limb amputees, each containing up to 400 images and videos, for a total of 800 pieces of content. Contributors include Ottobock ambassadors and users from across the world, representing different ages, ethnicities, abilities, and life contexts.

Once complete, the dataset will be published on HuggingFace – a widely used open-source platform for AI datasets – where developers, researchers, and AI teams can access and use it freely.

Current AI tools struggle significantly with representing people living with limb loss or limb differences. The most common failures are: erasure – the disability simply disappears from the generated image; incorrect depiction – a prosthetic limb is shown inaccurately or in the wrong context; stereotyping – images default to medical or inspirational clichés rather than everyday life; and in some cases, outright refusal – the AI declines to generate the image at all. One of our community members, Andreas Onea, tried to generate an AI image of himself as a superhero. The model kept giving him a partial arm. He has none.

The Dear AI library will improve this by giving AI systems real, annotated visual references created by the community itself – images of amputees cooking, working, dancing, parenting, playing sport, dating, and simply living. Not as symbols of disability, but as full human beings.

The current phase of the library is built by a curated group of Ottobock ambassadors and community members. However, the dataset will be published as open-source once complete – meaning anyone can access it, use it, and build on it. Future expansions of the library may include broader community contributions.

Nothing. The dataset will be freely available to everyone – companies, developers, researchers, educators, and individuals. There is no cost to access or use it.

The library will be published on HuggingFace – a leading open-source platform used widely by AI researchers and developers worldwide. Think of it as a public library for AI data: anyone can walk in, browse the collection, and take what they need. The dataset will be available with clear usage terms defined by Ottobock and the community, ensuring responsible use. The expected publication date is June 2026.

At Ottobock, as well as within the community, there is a shared understanding that representation of people with disabilities urgently needs improvement.

Currently, AI often represents people with disabilities inaccurately, stereotypically, or not at all. We have repeatedly observed that these portrayals are often highly unsatisfactory and far removed from the real-life experiences of people with disabilities. As a result, prejudices and misrepresentations continue to be reproduced and become further ingrained in people’s minds - and this needs to change.

The “Dear AI” initiative was created as a response to this issue. Together with creators from the community, Ottobock and Microsoft are developing a new, more authentic dataset for AI systems - one based on real experiences, real bodies, and real lives. The initiative consists of two key components:

On the one hand, a global campaign designed to raise awareness of the problem and find creative ways to bring attention to it: including personal letters addressed to AI from Ottobock Ambassadors, expressed through social media films, posters, projections, and more. On the other hand, it offers a concrete solution: the Community Library by Ottobock and Microsoft that helps improving representation.

The “Dear AI” initiative aims to make artificial intelligence more inclusive.

To achieve this, and to make representation tangible, Ottobock partnered with Microsoft. Together, they used the Community Library Maker - a platform that functions like a collaborative workshop where the community actively participates. Members select images, define themes, and teach AI what good representation actually looks like: from within the community, not from an external perspective.

This library is not created in isolation. It is built through conversations, diverse perspectives, and shared decision-making. Creators, users, and community members come together to define what feels authentic, what feels right, and what needs to change. Technology provides the tool - the community defines the truth.

The library is open source and free to download. Whether you are a developer, researcher, designer, or creative, you can use it to train or fine-tune AI models to generate more accurate, more human, and more inclusive representations of people with limb differences.