Thinkink: 2D Spatial Ink-native Interaction with LLMs

Ideation often involves externalizing thoughts through handwriting and sketching to explore concepts spatially. While Large Language Models (LLMs) can assist in brainstorming, current interfaces are typically text-centric and conversational, forcing users to switch between their creative canvas and a separate chat window. This disconnect disrupts the non-linear, visual nature of the ideation process. The authors developed Thinkink to bridge this gap, treating the digital canvas as a shared space where both the human and the AI contribute using ink.

The researchers followed a three-stage design process. A formative study identified that users value fluid, non-linear workflows and prefer minimal, paper-like interfaces. A subsequent diagnostic study with a technical probe revealed that users struggle with managing AI interactions without clear boundaries.

Thinkink addresses these challenges through a state-machine UI that separates note-taking from AI assistance. It uses a semantic tree to interpret the user's ink—distinguishing between drawings, concepts, and requests—and provides context-aware responses. These responses appear directly on the canvas as ink-like text or sketches, allowing the user to maintain their focus on the visual arrangement of ideas rather than managing a chatbot.

The final study with 10 participants demonstrated that Thinkink successfully supports an integrated ideation workflow. Users utilized the tool in four distinct ways: as a sounding board for reflective questions, as a generator for visual concepts, as a partner for proofreading and refinement, and as a source for expanding the breadth of their initial ideas. The design allows users to maintain agency over their creative process, using the AI to provoke new thoughts rather than simply providing final answers.