You give the generator five decisions and one optional sentence. The occasion tells it the dress code and the social stakes — a wedding guest outfit and a night-out outfit share almost nothing even when both are "dressy." The style direction tells it whose closet to raid: minimal keeps lines clean and logos absent, old money leans into tailoring and quiet fabrics, streetwear plays with proportion and layering, cottagecore softens everything into florals and natural textures. Season and color palette then constrain the fabrics and the mood, and the extra-details box is where you add the constraints that make the result yours — "no heels," "include the leather jacket I already own," "outdoor venue in October."
The output is a single full-body editorial photograph of the complete look on a clean studio background. That format is deliberate. A text list of garment names ("beige trousers, white shirt, loafers") leaves all the hard work — proportions, how the colors sit next to each other, what the silhouette actually does — to your imagination. An image settles those questions at a glance, which is why fashion magazines and e-commerce sites communicate in photographs rather than lists.
Each generation costs one credit, and regenerating with the same settings gives you a genuinely different take rather than the same outfit re-rendered — useful when the first result is close but the jacket is wrong. Small changes to the settings move the result a lot: the same "first date, classic, autumn" brief in neutrals versus jewel tones produces two outfits with completely different temperatures.