Please note: This master’s thesis presentation will take place in DC 2310.
Cara Li, Master’s candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Jian Zhao
Designers face increasing challenges when translating high-level creative intent into precise graphics operations. Traditional authoring tools expose low-level primitives that require complex sequences of tool invocations, imposing steep learning curves on users. While AI-assisted systems can interpret high-level intent, existing tools primarily support single-shot generation or style transfer, lacking scaffolding for iterative refinement and semantic-level reuse.
We present Semantic Ink, an interactive graphics authoring system built on three design principles: intent directness, where users express intent through freehand annotations directly on the canvas; steerability, where the system generates structured parameter spaces for progressive refinement; and reusability, where successful edits can be extracted as custom tools. Informed by a formative study with five designers, we contribute a two-dimensional taxonomy crossing five intent types (add, transform, style, remove, scene) with four controllable atoms (variation, magnitude, spatialization, surface), enabling multimodal LLMs to parse ambiguous annotations into editable parameter spaces. A user study with four experts demonstrates that Semantic Ink improves editing efficiency, reduces cognitive load from mode-switching, and enhances personalization through semantic-level tool reuse.