Enviar aquest missatge de text: AnySynth: Harnessing the Power of Image Synthetic Data Generation for Generalized Vision-Language Tasks