Motion grounded drag editing improves precision and realism in image manipulation
Drag as Evidence: Motion-Grounded Latent Recomposition for Drag-Based Editing
Computer Vision and Pattern Recognition
Summary
Editing images by dragging parts of them is simple but often struggles to be accurate while looking natural. The authors propose a new method called MoRe-Drag that uses the dragged motion as a guide inside a smart image generation process. This method gradually builds the changed image by mixing areas that need fixing or keeping and refines it step-by-step. The authors show MoRe-Drag works better than existing methods in precision and visual consistency, and it can even interpret drag gestures without extra instructions.
What this means in practice
- •For graphic designers: Improve precision and natural appearance when interactively repositioning image parts using drag motions in editing software.
- •For mobile app developers: Create user-friendly drag editing features for mobile photo apps that automatically understand and refine user adjustments without text input.$Commercial implications: Enables development of intuitive drag-based photo editing apps that provide high-quality, realistic edits, appealing to a wide consumer market.
Authors
Xinyu Pu, Hongsong Wang, Jie Gui, Pan Zhou
Abstract
Modern image editors excel at semantic manipulation and visual synthesis, yet remain limited in precise spatial control, motivating the development of drag-based editing. However, existing drag-based methods often struggle to balance drag accuracy with natural, plausible, and intent-aligned generation. We propose MoRe-Drag, a motion-grounded drag-based editing method. Our key insight is to treat pixel-space warping as coarse motion evidence, and to inject this evidence into the generative sampling trajectory. Specifically, MoRe-Drag performs region-aware latent recomposition over refinement, inpainting, and anchor regions, coupled with stage-adaptive conditioning that progressively shifts from motion-grounded structure formation to semantic refinement. We further support an instruction-free interface by adapting the MLLM-based text encoder for drag-aware instruction inference. Experiments on DragBench-SR and DragBench-DR show that MoRe-Drag substantially improves drag precision over strong base editors and achieves superior drag accuracy among SOTA drag-based methods, while delivering strong semantic consistency and visually realistic results. Code and dataset will be publicly released.