The pictures you add to a video now go to the model exactly as you uploaded them, with no resizing and no new compression; only location and camera details are removed. A picture is changed only where a model needs it: Seedance scales a picture longer than 2048 px down to that size, Sora fits it to the video frame, and a format a model does not take is converted. In the video studio on the web, a line under the pictures says which of these applies to the model you chose.