Training-free Geometric Image Editing on Diffusion Models

TL;DR

This paper introduces a training-free diffusion-based method called FreeFine for geometric image editing, effectively handling complex transformations by decoupling editing steps, and demonstrating superior performance on a new benchmark.

Contribution

It presents a novel decoupled pipeline for geometric editing that separates transformation, inpainting, and refinement, using training-free diffusion models, advancing the state-of-the-art in complex image edits.

Findings

Outperforms existing methods in image fidelity and edit precision

Effective on both 2D and 3D editing scenarios

Handles large and complex transformations successfully

Abstract

We tackle the task of geometric image editing, where an object within an image is repositioned, reoriented, or reshaped while preserving overall scene coherence. Previous diffusion-based editing methods often attempt to handle all relevant subtasks in a single step, proving difficult when transformations become large or structurally complex. We address this by proposing a decoupled pipeline that separates object transformation, source region inpainting, and target region refinement. Both inpainting and refinement are implemented using a training-free diffusion approach, FreeFine. In experiments on our new GeoBench benchmark, which contains both 2D and 3D editing scenarios, FreeFine outperforms state-of-the-art alternatives in image fidelity, and edit precision, especially under demanding transformations. Code and benchmark are available at: https://github.com/CIawevy/FreeFine