Rethinking Where to Edit: Task-Aware Localization for Instruction-Based Image Editing

Instruction-based image editing often causes unintended changes by over-editing. This work introduces a training-free framework that identifies precise edit regions using task-aware masks derived from stream-specific latent features. By guiding latent updates with these masks, the method ensures localized edits while preserving content consistency in non-edited areas.

Open the full episode

More from AI Radio