Automating hyperparameter selection for FlowEdit integration in Match-and-Fuse

Develop an automatic procedure for selecting FlowEdit hyperparameters within the combined Match-and-Fuse plus FlowEdit pipeline to achieve the desired balance between structure preservation and appearance change during localized editing of image sets, thereby eliminating the need for per-edit manual tuning.

Background

The paper integrates FlowEdit—an inversion-free text-based editing method—into Match-and-Fuse to enable localized edits that preserve background and structure while changing appearance according to prompts. Although default integration parameters are provided, the authors note that achieving the desired balance between structural preservation and appearance change often necessitates per-edit hyperparameter tuning.

This dependency on manual tuning motivates an explicit open problem: automatically determining FlowEdit’s integration hyperparameters (e.g., start timestep and control strength) for each edit so that the combined pipeline yields consistent, high-quality localized edits across diverse image sets without human intervention.