Reverie: A Gaussian Splat Study
A Gaussian splat study reconstructing runway footage into three-dimensional space.
Developed by Fashion Innovation Agency in collaboration with Tamaris Ellis, Reverie is a Gaussian splat R&D study exploring monocular image synthesis as a new method for representing fashion in three-dimensional space. Working entirely from found footage, the project investigates how spatial depth can be inferred from a single camera angle, without a studio, rig, or 3D scan.
Reverie reframes found runway footage as spatial material, transforming a linear sequence of images into a navigable three-dimensional environment shaped through custom code, AI inference, and intensive post-production.
Reverie reconstructs 23 looks from the 2023 LCF MA graduate show into volumetric three-dimensional space from a single fixed camera angle. Each frame of the original runway footage is processed individually through a monocular Gaussian splatting inference model, rebuilt in three dimensions, then reassembled into a continuous spatial sequence by a bespoke pipeline built entirely from scratch. The output is a study where garments and figures exist as point clouds: bodies that materialise out of particles, move through space, and dissolve into themselves on exit. Painterly, sculptural, and built entirely from what was already there.
Using monocular image synthesis, custom scripting and Gaussian splatting, Reverie transforms found runway footage into a navigable three-dimensional fashion study.
Using monocular image synthesis and a bespoke pipeline built from scratch, Reverie investigates how found footage can be transformed into new forms of three-dimensional spatial expression. Runway looks are broken down frame by frame, volumetrically rebuilt through AI inference, and reassembled into a continuous sequence where garments and figures exist as sculpted point clouds. Through extensive custom scripting and post-production, the project reconciles the fundamental instability of per-frame reconstruction into a coherent moving study. Reverie explores how emerging spatial rendering techniques can represent fashion in three dimensions without modelling, simulation, or recreation from scratch.