Centre for informatics and computing at Ruđer Bošković Institute (CIR) participated in the Technical Culture Days, held on 18 and 19 September 2026 at the Nikola Tesla Technical Museum in Zagreb. The event brought together numerous organizations, associations, and institutions from the fields of technology, science, and STEM, offering visitors a wide range of workshops, demonstrations, and interactive activities.

As one of the participants, CIR introduced concepts of high-performance, parallel, and distributed computing trough an interactive approach encouraging children and young people to explore, learn, and engage in creative activities.

The concept was simple: children draw an imaginary creature on a tablet, and AI transforms their sketch into a colourful character inspired by the visual style of a 3D animated film. The newly created character then joins others in a shared “Friends' Room” displayed on a second screen, where children can move and arrange their characters together. As a keepsake, each child also received a print featuring their original drawing alongside its AI-generated interpretation.

A new character was generated in approximately 30 seconds. Rather than displaying a conventional loading indicator, we let children watch the image gradually emerge from visual noise, providing an intuitive introduction to the basic idea behind diffusion-based image generation. This also creates an opportunity to discuss how the original drawing influences the result, what the AI contributes, and where the computation takes place. Importantly, the entire process ran locally, without an internet connection.

What was used:

🔹 Stable Diffusion 1.5 + ControlNet Scribble — the child's sketch guides the character's form, while a text prompt influences its style, textures, and lighting.

🔹 LCM-LoRA — reduces the number of generation steps, allowing intermediate results to be displayed in near real time.

🔹 Rembg — removes the generated character's background before placing it in the shared “Friends' Room.”

🔹 Electron + React — powers the two-screen user interface.

🔹 Python + FastAPI — handles the local AI inference pipeline.