Work

Generative Scaffolding for Non-Visual Creative Intent: Supporting BLV Creatorsin AI Image Generation

Year: 2026
Type: dissertation
Abstract: Visual content creation has become a primary mode of social media participation, and AI image generation tools have theoretically opened visual content creation to blind and low vision (BLV) creators by replacing visual execution with language input. However, existing tools assume that users can independently view, evaluate, and refine generated outputs, an assumption that systematically excludes BLV creators from independent visual content creation. This study adopts a Research through Design approach to investigate how AI tools can be designed to help BLV creators generate images that match their intent. Through formative interviews with 12 BLV creators, the study identifies disruptions in the intent-to-outcome chain across the AI image generation workflow and examines how BLV creators conceptualize and articulate visual ideas through tactile, spatial, emotional, semantic, and remembered visual resources. Based on these findings, three design considerations were derived and used to develop a conversational generative scaffolding prototype. The prototype was evaluated with 6 BLV creators, confirming its effectiveness at the conception and execution stages while revealing a structural boundary at verification. The findings show that scaffolding effectiveness depends on the richness of intent a creator brings into the conversation, and that the absence of a reliable shared perceptual reference between non-visual cognition and visual expression constrains both intent transmission and result verification. At the theoretical level, the study proposes that sense of process control and sense of ownership within creative agency do not necessarily align in conversational AI collaboration.
Author: Zhang, Tan
Language: English
Open Access status: green
OpenAlex ID: W7165989343
Cites:
Cited by:
Related to: 10