Point2pix-Zero: Point-driven refined diffusion for multi-object image editing

Open Webpage

Siyuan Wang, Yuyao Yan, Xi Yang, Rui Zhang, Qiufeng Wang, Guangliang Cheng, Kaizhu Huang

Published: 01 Jan 2026, Last Modified: 09 Nov 2025Pattern Recognit. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We propose a novel point-driven framework for inference-time image editing.•Our method edits objects from a few points, automatically identifying visual and semantic information.•We introduce a refined diffusion strategy to mitigate visual and semantic mismatches during editing.•Extensive experiments on single and multi-object images demonstrate our approach’s efficacy.