Tag
This paper presents MAOAM, a unified vision-language model framework that enables precise object and material selection through text or click interactions for interactive image editing. It introduces a scalable data generation pipeline and shows emergent improvement when combining text and clicks at inference.