Back to Projects
HUMAN IN THE LOOPMultimodal installation, 2025
Role
Concept, 3D Design & Visuals
Team
3 person Collaboration
Format AI-assisted Installation
Tools ComfyUI,Diffusion Models,Blender
Year 2025
A speculative experiment where human cognition and robotic AI jointly generate string figure patterns. By mixing human body movement, mechanical precision, and algorithmic inference, the project exposes the cognitive gap between human spatial intuition and machine logic. It highlights the disharmony of human machine collaboration, presenting a new dynamic defined not by perfection, but by uncertainty, failure, and computational hallucination.
INPUT
Serving as the primary input for the entire workflow, the unconscious manual play of string figures is used to explore the machine's comprehension of uniquely human actions.
OUTCOME
The work focuses on the attempts, failures and unstable structures produced during the experiment rather than a polished result. Unable to truly understand the image, the AI reconstructs the string figures through its own topological logic, producing computational hallucinations that reveal the gap between human spatial intuition and algorithmic construction.
COGNITIVE TRANSLATION
1. Frame Capture (Gradio)Extracts the final frame from the string-performance video.
2. Algorithmic Misreading (ComfyUI & Qwen-Edit)
Reinterprets the captured form through diffusion-based image editing.
3. Computational Hallucination
Translates fluid string forms into a machine-generated visual language.
Phase 1 — Noise: Random noise establishes the image’s rough structure, texture and lighting.
Phase 2 — Refinement: The algorithm gradually removes noise, identifying lines and directional flow to clarify the forms.
Phase 3 — Emergence: Like a photograph developing in a darkroom, the final image gradually emerges from the noise.
Back to Projects