Abstract
A point-and-describe image editing technique captures pointer-based input on a canvas or workspace together with voice annotation and optional scene-level context. The pointer input can identify two-dimensional or three-dimensional coordinates, and speech-to-text processing can convert spoken objects or edit descriptions into text. The coordinates can be mapped to natural spatial descriptors, normalized coordinate metadata, or spatial mask data. Before submitting visual input for image generation or localized image editing, the system can flatten drawing strokes while hiding visible text labels, then package the clean canvas with spatial prompt metadata. This supports faster multimodal image generation and localized editing in keyboard-limited computing environments.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Lane, Michael, "Point-and-Describe Spatial Multimodal Image Editing and Generation", Technical Disclosure Commons, (November 10, 2026)
https://www.tdcommons.org/dpubs_series/12948