Three garments, one word
A saree is six yards of unstitched cloth, a blouse and a petticoat. A lehenga is a skirt, a blouse and a dupatta. A salwar suit is a tunic, a lower garment and a dupatta.
An image model reading any of those as a single object hands back the western garment they most resemble, which is a gown, sometimes with a scarf draped over one shoulder as a nod to the input.
The prior is not wrong, it is just larger
The model has seen far more evening gowns than sarees. Absent an instruction that outranks the resemblance, it goes with the shape it knows. That is not a failure of capability. Ask it plainly enough and it draws a correct saree.
Plainly enough is the operative phrase, and it means naming the construction rather than the garment.
What a silhouette rule contains
Ten ethnic categories here carry their own rule. The saree rule states the drape: unstitched cloth wrapped around the waist, pleated at the front, the loose end carried over one shoulder, worn with a fitted blouse ending above the waist.
The sherwani rule states that it is a long structured buttoned coat over churidar or trousers, and stays one. The kurta rule fixes where it ends, between hip and calf, over separate lower wear.
These hold regardless of which model is cast, which is the point. A rule attached to the category cannot be forgotten by whoever fills in the form.
What still comes from your photograph
The silhouette rule fixes the construction. Everything specific to your piece, the colour, the print layout, the zari, the mirror work, the border, the trims, is carried from the reference photograph and preserved.
That division is worth stating because it is what makes the output your garment rather than a plausible one. The model generates the person, the room and the light. It does not get to redesign the clothes.