Dreams of Blindness: An AI-Mediated Workflow

Abstract

Dreams of Blindness is a VR documentary that explores dream experiences narrated by visually impaired people. Drawing from more than twenty interviews, five dream narratives were selected and reconstructed as immersive scenes through an AI-mediated workflow connecting semantic interpretation, image and 3D generation, motion, sound, and spatial composition. Generated material was treated as provisional rather than authoritative: creators repeatedly selected, rejected, and revised outputs according to the emotional and experiential logic of the source accounts. Abstract forms, point clouds, diffuse light, spatial sound, and motion supported the translation of non-visual sensations without claiming to reproduce another person’s perception. The project positions human judgment and representational responsibility at the center of AI-assisted documentary practice.

Original Figures

Workflow connecting dream narratives, AI-mediated generation, iterative human evaluation, and VR scene construction. — The workflow translates dream narratives through semantic mediation, generative modeling, and spatial composition into VR scenes. Human evaluation remains iterative throughout, allowing creators to select, reject, and refine generated material.

Example showing rejection of an unsupported stereotyped detail and revision toward an abstract representation aligned with the source account. — This example follows a narrative extract from an initial AI image containing an unsupported stereotypical detail to a more abstract final representation. Human intervention retained a meaningful visual cue while rejecting and transforming elements that did not fit the source account.

Publication

Jiang, L. (2026). Dreams of blindness: An AI-mediated workflow for creative representation in VR documentary. In Proceedings of the 2026 ACM Conference on Creativity and Cognition (pp. 1651–1653). Association for Computing Machinery. https://doi.org/10.1145/3803784.3816877