Transferscope — Synthesized Reality

What if the environment itself could become a prompt? Instead of searching for the right words or relying on a keyboard, the act of looking + capturing becomes the starting point for creation.

During my residency (Feb–June 2023) at the AI+D Lab at HfG Schwäbisch Gmünd, I explored generative image synthesis through Stable Diffusion. My interest centered on shifting the interaction away from text-based prompting toward something more embodied—using a physical tool to mediate between the user, the environment, and the model. This investigation led to the development of Transferscope.

Transferscope is a handheld device that offers a tangible way of engaging with generative AI. Built with open hardware, it invites others to construct, adapt, and repurpose it. A camera captures the scene in front of the device, which is then interpreted by a generative model and returned as a transformed image on the integrated display. The process feels immediate and intuitive, allowing users to sample and remix reality through an AI-driven lens.

Rather than representing today’s highly refined image-generation systems, Transferscope captures a specific technological moment. It reflects the state of generative image tools as they existed in 2023—less polished, more unpredictable, and defined by the small distortions and ambiguities characteristic of early Stable Diffusion models like 1.5. These imperfections are central to the project. They introduce uncertainty, invite experimentation, and give each transformation a sense of openness. Transferscope preserves this raw phase of AI image-making, a time when the technology’s creative potential was intertwined with its errors.

Despite the complexity of the underlying systems, the interaction is intentionally simple. This contrast highlights a core tension in contemporary AI design: advanced computation made accessible through a familiar and almost playful interface. Transferscope underscores the collaborative potential of AI, framing it as a creative partner rather than a black-box generator.

The project developed through an iterative process of prototyping and refinement. Several 3D‑printed models were tested, evaluated, and reworked. It builds on ideas explored in an earlier project, unStable Mirror, and eventually consolidated what began as two separate devices into a unified object with a clearer identity and experience.

As generative image technologies continue to evolve, Transferscope remains an ongoing inquiry into how we might engage with AI in ways that feel grounded, material, and situated in the world rather than abstracted behind a screen.

I am grateful to the AI+D Lab team: Benedikt Groß, Aeneas Stankowski, Rahel Flechtner, Felix Sewing, and Stamatia Galanis—for their support, insights, and patience throughout this process.

Materials: 3D‑printed handheld device with Raspberry Pi, Linux, Python, OpenCV, ComfyUI, ControlNet, IPAdapter, and Stable Diffusion.