What would your feature do ?
Right now if you have an image input and you have auto-detect on + one of auto-fit/auto-resize, they get resized but they DONT respect the grid steps. So you end up with very odd numbers, which sometimes might not work (depending on what model/attention you're using).
Proposed workflow
I propose to have an additional toggle for auto-snap (after all the other toggles)
Additional information
I can contribute the PR for this if you want.