✦
Summary
SpatialEdit is an open-source benchmark and research resource for fine-grained image spatial editing. It is designed to test whether image-editing models can follow precise spatial instructions, such as moving, resizing, rotating, or repositioning visual elements while preserving the rest of the scene. This makes it important for evaluating whether image editing systems are controllable enough for real design and production workflows.
Key Features
Benchmarks fine-grained spatial image editing.
Tests precise operations such as moving or repositioning content.
Measures instruction following and preservation together.
Useful for evaluating diffusion-based image editors.
Supports research into controllable visual editing.
Provides public code and benchmark materials.
Helps compare models beyond general aesthetic quality.
Relevant to design tools and production image workflows.
About SpatialEdit
The benchmark focuses on spatial edit operations and likely pairs input images with structured instructions and target outcomes. Technically, this kind of evaluation requires checking both instruction adherence and preservation: the edited object must change in the requested spatial way, while unrelated content should remain stable. That is harder than broad style transfer or global image modification.
SpatialEdit is valuable for researchers and developers building controllable image editors, generative design tools, and diffusion-based editing pipelines. It gives teams a way to measure the spatial precision of their systems instead of relying only on visual appeal.