Bulk background remover

Ten minutes of tidying before the upload saves an hour of sorting afterwards. Name the files for what they show, by product code or by person, because the results keep those names and a ZIP full of IMG_4471 to IMG_4490 tells you nothing. Remove near-identical shots, so you are not paying attention to the same picture three times. Check sizes as well: each file may be up to 20 MB on the free plan and up to 25 megapixels, and an image over the limit is refused while its neighbours go through.

If the pictures still have to be taken, keep the camera position and the light the same for the whole series. A model that sees twenty similar photos makes twenty similar decisions, and the batch comes out even.

The model handles one image at a time, a few seconds each, so twenty pictures are done in a minute or two and several hundred take long enough to fetch a coffee. Leave the tab open while it runs. The progress count tells you where it stands, and finished pictures can be inspected while the others are still waiting.

The slow part is finding the subject, and recent cut-outs are remembered for a while. If you decide at the end that the range should be on light grey instead of white, the pictures processed most recently are repainted from the masks that already exist, which is much faster than the first run. In a very large batch the oldest ones are worked out again.

Nobody checks five hundred edges one by one, and nobody needs to. Scroll the thumbnails first: the failures that matter are obvious at that size, a missing limb, a product cut in half, a patch of background left standing. Open only those and compare them with the divider. For the remaining pictures, spot-check a handful in full. Images that came out wrong usually share a cause, such as a subject that touches the edge of the frame, and are quicker to reshoot or crop than to rescue.