Pošalji tekstualnu poruku: AnySynth: Harnessing the Power of Image Synthetic Data Generation for Generalized Vision-Language Tasks