A Local Product Photo Pipeline: Cut Out, Upscale, Compress, Ship

Listing photos are the highest-leverage, lowest-glamour part of selling online. The same product photographed identically will convert differently depending on whether the background is clean, whether the image loads instantly, and whether every photo in your catalogue looks like it came from the same shoot. Agencies charge for all three.

You can do the whole chain in a browser tab, on a folder of two hundred images, for nothing. Here is the order that works and why.

Order matters: cut out before you compress

The single most common mistake is compressing first. Once you have discarded detail, every later operation works from degraded data: background removal finds harder edges, upscaling amplifies artefacts, and a re-compress at the end stacks another generation of loss on top.

The correct order is remove background → repair/upscale → compose → compress once, last.

Step 1: remove the background

Load the whole folder in batch mode. For products rather than people, use the general-purpose tier; the portrait-optimised model is faster but it is tuned for faces and will treat a handbag accordingly.

Then set the finishing options once and let them apply to the batch:

• Solid white background if your marketplace requires it; transparent if you are going into design work.
• 1:1 aspect for marketplace grids; 9:16 only where stories or vertical placements demand it.
• Crop to subject — trimming the dead space so the product fills the frame. This is what makes a catalogue look consistent even when the source photos were not.
• A faint contact shadow. Products cut out and dropped on pure white read as floating stickers. A soft shadow reads as photographed.

Step 2: rescue the old stock

Every catalogue has legacy images — a 600-pixel photo from a supplier, a JPEG that has been re-saved five times. Two tools apply, in this order:

• Restore first if the source is damaged: scratches, JPEG blocking, noise, faded colour. Repairing before enlarging stops you upscaling artefacts into permanent features.
• Upscale second to reach your target size. Be realistic: 2× is a reliable lift, 4× is a genuine rescue for small sources and a disappointment if you expect invented detail.

Crucially, upscaling happens before the final compression andafter restoration, so you compress exactly once.

Step 3: compose, then compress once

With the cutouts at full quality, compose the final frame — background, aspect, shadow — and only then run compression. Targets that hold up in practice:

• WebP or AVIF at quality 70–80 for photographs. AVIF is materially smaller; WebP has broader support. Serve AVIF with a WebP or JPEG fallback where you control the page.
• Under ~200 KB per listing image is an achievable target for a 1000–1600 px product shot and keeps category pages fast on mobile.
• Strip metadata. EXIF from a camera can carry GPS coordinates and device serials. Removing it is smaller and private.
• Keep a lossless master. Compress the copy you publish; never overwrite the original.

Step 4: export in one go

Batch export as a ZIP gives you a single file to drop into your store admin. Because nothing is uploaded, there is no queue, no per-image credit, and no size cap — the only limit is your machine’s patience, and the queue keeps memory flat by holding thumbnails rather than hundreds of full-resolution canvases.

What this replaces

The commercial version of this pipeline bills per background removal, caps the free tier at low resolution, queues batches, and holds your images on its infrastructure. For a small catalogue that is an annoyance; for a large one it is a monthly line item, and for pre-release product photography it is a confidentiality problem — those images are not public yet.

Running locally means unlaunched products stay unlaunched. Start with the background remover in batch mode, then finish the folder through the compressor.