After recent deve­lo­p­ments in GenAI, par­ti­cu­lar­ly in Visi­on-Lan­­guage Models (VLMs), the fashion pro­ject I star­ted 1.5 years ago to deve­lop an out­fit recom­men­da­ti­on sys­tem seems alre­a­dy out­da­ted. The pro­ject is out­lined here. In ear­ly 2023, using VLMs was alre­a­dy an opti­on; howe­ver, due to the insuf­fi­ci­ent matu­ri­ty of available models, I cho­se a dif­fe­rent approach descri­bed here. Dri­ven by curio­si­ty, I deci­ded to revi­sit the pro­ject in a redu­ced form, as sca­la­bi­li­ty is not a con­cern for this ite­ra­ti­on.

Fashion is high­ly sub­jec­ti­ve, making it dif­fi­cult to deve­lop a sys­tem that can uni­ver­sal­ly recom­mend sty­lish and diver­se out­fits. Indi­vi­du­al pre­fe­ren­ces, sar­to­ri­al liter­acy, and per­so­nal expe­ri­en­ces great­ly influence how peo­p­le per­cei­ve and cate­go­ri­ze styl­es. Deve­lo­ping an algo­rithm that can navi­ga­te the­se nuan­ces and deli­ver per­so­na­li­zed, accu­ra­te recom­men­da­ti­ons is a com­plex task.