Conceptual
Browse Instruct Curate

Diffusion-Based Image Editing with Attention Control

Techniques that repurpose a pretrained text-to-image diffusion model to edit an existing image rather than generate one from scratch, typically by manipulating the model's cross-attention maps (as in Prompt-to-Prompt and Plug-and-Play) or by instruction-tuning it on edit pairs (as in InstructPix2Pix). The goal is precise, localized changes that preserve the unedited structure of the source image.