Achieves zero-shot, prompt-free object removal in diffusion models purely through self-attention manipulation.
March 31, 2026
Original Paper
PANDORA: Pixel-wise Attention Dissolution and Latent Guidance for Zero-Shot Object Removal
arXiv · 2603.27555
The Takeaway
Unlike previous methods requiring fine-tuning or complex prompts, PANDORA uses 'Attention Dissolution' to erase objects by nullifying specific attention keys. This allows for scalable, non-rigid object erasure in a single pass using off-the-shelf pre-trained models.
From the abstract
Removing objects from natural images is challenging due to difficulty of synthesizing semantically coherent content while preserving background integrity. Existing methods often rely on fine-tuning, prompt engineering, or inference-time optimization, yet still suffer from texture inconsistency, rigid artifacts, weak foreground-background disentanglement, and poor scalability for multi-object removal. We propose a novel zero-shot object removal framework, namely PANDORA, that operates directly on