Multi-turn image editing dataset improves AI editing quality across steps
MTEdit: Packed Trajectories Beat Rollouts for Multi-Turn Image Editing
Computer Vision and Pattern Recognition
Summary
Editing images step-by-step with AI can cause the image to lose quality or drift away from the original intent after several changes. The authors created a large dataset called MTEdit with many sequences of editing steps designed to keep the image quality high and the edits consistent. They tested this dataset by training an AI model and showed it performs better at following instructions and keeping edits clear over multiple turns. The dataset and benchmark allow better development of AI that works well with people to edit images interactively.
What this means in practice
- •For graphic design teams: Create AI-assisted image editing tools that maintain high visual quality across multiple sequential edits based on user instructions.
- •For software developers: Integrate multi-turn editing models trained on MTEdit to improve interactive image editing features in creative software.
Authors
Houyuan Chen, Fanzhou Wang, Xiangyu Fan, Songlin Yang, Zhitao Yang, Lvmin Zhang, Pei Yang, Hao Zhao, Lei Yang, Anyi Rao
Abstract
Multi-turn image editing aims to modify an image according to a sequence of instructions, making it better aligned with human-AI collaborative creation than single-turn editing. However, constructing high-quality multi-turn training data remains challenging, as conventional sequential rollout accumulates errors and introduces visual-quality drift across editing turns. To address this challenge, we introduce MTEdit, a multi-turn image editing dataset built through a pipeline centered on packed trajectory canvas (PTC) generation. By jointly generating multiple editing states within a unified visual context, this pipeline reduces visual-quality drift across turns. Covering both step-wise and context-aware editing scenarios, MTEdit comprises 500K multi-turn image editing trajectories, with up to 8 turns and an average of 5.38 turns per trajectory, making it one of the most extensive datasets for multi-turn image editing. We further introduce MTEdit-Bench, a benchmark designed to evaluate both scenarios by assessing models' ability to follow editing instructions, maintain content consistency, resist visual-quality drift, and effectively use historical context across turns. Using MTEdit, we fine-tune Qwen-Image-Edit-Causal to obtain Qwen-Image-MTEdit. On MTEdit-Bench, we evaluate a broad range of closed-source and open-source image editing models, including Qwen-Image-MTEdit. Experimental results show that Qwen-Image-MTEdit achieves the best performance among the evaluated open-source models, demonstrating the effectiveness of MTEdit in improving models' multi-turn image editing capabilities.