Bulk background removal: processing a whole catalogue without losing a week

Consistency is the actual goal

Nobody notices that image 212 is slightly better than image 213. Everybody notices when the products on a category page are shot at different distances, lit from different sides, and sitting at different heights in the frame. A catalogue reads as professional or amateur as a whole, not image by image.

That means the useful question is not "how do I get the best cutout of this item" but "how do I get an acceptable cutout of every item in the same way". Automation is what makes the second question answerable.

Fix the shoot before you fix the photos

• One position, one camera height, one distance. Mark the floor with tape if you have to.
• One background, plain and evenly lit, in a colour that contrasts with most of your products. Mid-grey is a safer default than white if you sell light-coloured items.
• Same lens and focal length throughout; a zoom lens set to different lengths distorts products differently.
• Name files as you shoot, by SKU. Renaming four hundred images afterwards is the task that never gets done.
• Shoot the hardest items first — anything glass, reflective, white, or with fine detail — so you find out early whether the setup works for them.

Batching the processing

Upload in groups rather than all at once. Fifty at a time is a size where a systematic problem — the wrong background colour, a lighting change halfway through the shoot — shows up before you have processed the whole catalogue with it.

Decide the output format once. For a catalogue, transparent PNG as the master is almost always right: every marketplace version, every colour variant and every promotional composite derives from it, and you never have to re-process.

Keep the originals. Storage is cheap, and a re-shoot is not.

Reviewing four hundred results without looking at four hundred images

Do not review at thumbnail size on a white page; that hides exactly the problems you are looking for. Set the preview background to black or a saturated colour, where a halo, a missed shadow or a chopped edge is obvious.

Sort by product type and spot-check the categories that fail most: anything with holes (chairs, jewellery, handles), anything translucent or reflective, anything fluffy, and anything white. Solid opaque objects against a plain background almost never fail and do not need individual attention.

Budget for a manual pass on roughly five percent of the catalogue. That is the honest number. A tool that gets 95% right unattended has done its job; the remaining 5% is where a person adds value, and it is a far better use of a person than the other 95%.

When to move to an API

A bulk uploader is the right tool when a person is already looking at the images: a new range, a seasonal shoot, a marketplace migration. It stops being the right tool when images arrive continuously — supplier feeds, user uploads, a PIM system that ingests on a schedule.

At that point the processing should happen where the images already are, without anyone dragging files into a browser. An API call per image, triggered by whatever system receives them, means the catalogue is always clean and nobody has a recurring task on their list.

The threshold is not a number of images. It is whether the work is a project or a process. Projects suit a bulk uploader; processes suit an API.