1 paper
Jiajia Lin, Mingxuan Du, Tuowen Zhou +2
Text-to-image and personalized editing models now synthesize high-fidelity single-subject images with ease. Yet placing multiple named people into shared contact actions such as em…