VizXpress: Towards Expressive Visual Content by Blind Creators Through AI Support
Lotus Zhang, Zhuohao (Jerry) Zhang, Gina Clepper, Franklin Mingzhe Li, Patrick Carrington, Jacob O. Wobbrock, Leah Findlater · 2025 · ASSETS 2025: 27th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663547.3746345
Summary
This paper investigates how blind individuals engage with and desire to create expressive visual content — moving beyond the functional accessibility tasks (like privacy preservation or layout correction) that prior research has focused on. The authors conducted a two-stage study: first, interviews with 10 blind participants to understand their motivations, current practices, and barriers in visual expression; second, the development and evaluation of VizXpress, an interactive prototype that serves as a design probe for exploring accessible visual editing. VizXpress is a web-based image editing tool powered by GPT-4o that provides real-time AI-generated feedback on image aesthetics (clarity, framing, overall feeling, style, color, lighting), actionable editing suggestions, and a question-and-answer channel. It supports both AI-assisted automatic edits (where users describe their intent and the system generates editing recommendations) and manual screen-reader-friendly controls for color and lighting adjustment, filters, object-based cropping, text insertion, and sticker placement. The design probe study with 14 blind participants (including 4 from the initial interviews) explored how these tools supported expressive visual creation, revealing nuanced information needs around subjective aesthetics, challenges with perceiving subtle visual changes, and the tension between AI automation and personal creative control. The 20 unique participants across both studies were recruited through the National Federation of the Blind and represented a range of visual conditions, onset ages, and levels of visual memory.
Key findings
Blind individuals demonstrated strong, diverse interests in expressive visual creation spanning professional goals (teaching blogs, concert promotion, business branding), social sharing (social media aesthetics, greeting cards), and artistic pursuits (photography, illustration, meme-making, generative AI art). Most participants (8 of 10) had not used expressive editing features like filters or stickers due to inaccessible tools, relying instead on sighted helpers — which raised issues of independence, privacy, communication gaps, and creative agency. In the design probe study, participants valued the AI aesthetic feedback for discovering aspects of their images they could not independently perceive, but found it insufficient for verifying subtle color and lighting changes after edits. Eleven of 14 participants felt the feedback was too abstract for nuanced editing decisions. Participants strongly valued before-and-after comparisons to understand the impact of their edits. Object-based cropping was found highly intuitive (12 of 14 participants used it), while descriptive font and color labels (e.g., "Comic Sans: casual and playful") were praised as empowering. Participants wanted both AI automation for speed and manual controls for creative ownership — seeing them as complementary rather than competing approaches. Six participants raised concerns about AI feedback being overly positive or filtered, potentially masking honest aesthetic evaluations. All 14 participants expressed excitement about the potential for visual expression tools, with applications envisioned across personal branding, professional documents, social media, video editing, and generative art.
Relevance
This research represents an important shift in accessibility research from functional accommodation to creative empowerment, recognizing that blind individuals have aesthetic sensibilities and expressive desires that deserve technological support. For accessibility practitioners, the findings challenge the assumption that visual content creation is inherently a sighted activity, and highlight how AI can serve as a bridge — not by making decisions for blind creators, but by providing the perceptual information needed for informed creative choices. The design of VizXpress offers concrete patterns for accessible creative tools: descriptive labels that convey aesthetic qualities without requiring visual experience, object-based rather than position-based interactions, hierarchical feedback structures, and the combination of automated and manual editing controls. The study also raises critical questions about AI honesty and bias in aesthetic evaluation — findings relevant to anyone building AI-powered accessibility tools. The emphasis on creative agency and the right to aesthetic expression connects to broader disability rights frameworks and challenges the tendency in assistive technology to focus solely on functional tasks.
Tags: blindness · visual expression · creativity support · AI accessibility · image editing · vision-language model · screen reader · aesthetic feedback · design probe · generative AI