When you upload a product photo, the description is written for you from what the photo shows. Those descriptions were also recording facts about the photograph: that the product was seen from the back, laid flat, worn by a model, or shown close up. None of that describes the product itself — a bikini is not "shown from the back", it simply has a back.
This mattered beyond the wording. A product's description is passed to the video model as the description of the product, so a description saying the item was "shown from the back on a model" was effectively instructing the video model on how to frame the shot. Which way the product faces is already recorded separately, on the photo.
New descriptions now state only what is true of the product: its colours, materials, construction, detailing, and how many pieces it comprises. Facts about the photo go where they belong — the direction the product faces stays on the photo, and the setting goes into the background description.
Existing descriptions have been corrected too. Every affected description was rewritten to remove the photographic wording while keeping every product fact, and each rewrite was checked before being saved: anything that would have dropped a detail or introduced one that was not already there was left untouched for review instead.
Descriptions you have edited yourself are not affected, and a corrected description will not be overwritten by later analysis.