Product photo pipeline — cutout, white background, resize and compress in one pass
Getting a supplier snapshot ready for a marketplace normally means four tools, four downloads and four renames: remove the background, save, resize, save, compress, save. Every save re-encodes the file, and every step has to guess a number the next step actually needs. This page runs the whole chain on one canvas and reports exactly what it did to each file. Nothing is uploaded.
Works on studio sweeps and plain walls. Colour-based, so it is instant and predictable — and it cannot handle a busy background.
Square canvas, white padding, product centred at a set proportion of the frame.
Searches for the highest quality that still fits, and reduces pixel dimensions only when quality alone cannot get there.
Why four separate tools is the wrong workflow
The obvious cost of using four tools is time. The cost that actually shows up in the finished file is re-encoding. JPEG and WebP are lossy: every time the file is decoded and written back out, artefacts accumulate and detail is discarded permanently. A photo that goes background remover, then resize tool, then compressor has been through three lossy generations before a shopper ever sees it. This page composites everything on one canvas and encodes once.
The second cost is quieter and does more damage. Each step depends on a number the previous step knows and you do not. The resize step needs the product's actual size in the frame; the compressor needs to know the final pixel count, because file size scales with it. Run as separate tools and you are typing in estimates — and when the result is wrong there is no way to tell which of the four steps was responsible.
So the order here is fixed and not configurable: background, then frame, then file weight. Each stage hands the next one a measurement rather than an estimate.
What each step actually does
Step 1 — remove a flat background
The tool samples the outer band of the photo to find the dominant colour, then judges every pixel by how far it sits from that colour. Close to the backdrop a pixel becomes fully transparent; far from it, fully opaque; in between it becomes semi-transparent, which is what produces a natural edge rather than a staircase. Tolerance sets where that transition begins and ends.
This is a flat background remover, not an AI cutout, and the distinction is worth being blunt about. Because it measures colour instead of recognising objects, it is instant, needs no model download, and gives the same answer every time. It also only works when the backdrop really is one consistent colour. Studio sweeps and plain walls: excellent. Grey seamless with a gradient across it: usually fine. A room, a table, a patterned cloth: it will not work, and no tolerance value will fix it. That case needs a segmentation model, and those either cost money or require uploading your photos somewhere.
One constraint that catches people out, because it is invisible until it happens: the product has to sit inside the frame with backdrop visible around it. Detection works out the backdrop by taking a majority vote across the outer band of the photo, so if the product runs off all four edges the tool samples the product's own colour, calls that the background, and removes the subject instead. Leave a margin when you shoot. A product resting on the bottom edge of the frame is fine — three of the four edges still belong to the backdrop.
Why the trim matters more than people expect
Once the background is gone, the subject is sitting in a field of white with no fixed relationship to the frame. Measure it: a product photographed on a large sweep might occupy 40 percent of the width; a tightly framed shot of the same product might occupy 95 percent. There is no way to know in advance, and it varies from photo to photo in the same batch.
That is why “product fills 85 percent of the frame” only means something after the empty margin has been cropped away. Trimming to the subject's bounding box turns the product's size into a measured number, and only then can the frame proportion be applied consistently across a whole folder. Skipping this step is the single most common reason a batch of photos comes out with wildly different product sizes.
Step 2 — place it in a marketplace frame
A square canvas is filled with the pad colour, the trimmed product is scaled so its longest side takes the set proportion of the frame, and it is centred. Because the source was trimmed first, that proportion is honoured exactly: 85 percent means 85 percent on every photo in the batch, whatever shape the originals were.
Choosing fill frame and crop edges instead scales until the frame is covered and trims the overflow. Use it for lifestyle shots where the edges are expendable. Never for a product that touches the edge of the photo — you will cut it in half.
Step 3 — hit a file size
Rather than asking you to guess a quality value, this step searches for it. The encoder responds monotonically — lower quality always means a smaller file — so a binary search finds the highest quality that still fits within a handful of attempts. If no quality produces a small enough file, the tool works out how much the pixel dimensions need to shrink, resamples once and searches again.
There is a floor worth knowing about. Below roughly quality 30 to 40 on a JPEG, further reductions buy very little size and start destroying the image: blocking around every edge, banding in gradients, text turning to mush. Hitting a very small target is almost always better done by reducing dimensions than by crushing quality.
Marketplace frame sizes
| Platform | Main image | Notes |
|---|---|---|
| Amazon | 2000 × 2000 | 1000 px minimum for zoom, 1600 px commonly recommended; pure white background required on the main image; product roughly 85% of the frame |
| eBay | 1600 × 1600 | 500 px minimum, 1600 px for the zoom feature |
| Shopify | 2048 × 2048 | 2048 px square is the theme default for maximum zoom |
| Google Shopping | 1200 × 1200 | Images under 100 px are rejected outright; 800 px is the floor for large display formats |
| Shopee / Lazada | 1000 × 1000 | Square, minimum 500 px, white or light background |
| Taobao / Tmall | 800 × 800 | Square required |
If you sell on more than one platform, run the pipeline once per platform rather than hunting for a size that satisfies all of them. A single compromise size usually fails the strictest platform and wastes bandwidth on the rest.
The settings that decide quality
Tolerance
How different from the detected backdrop a pixel has to be before it counts as product. Too low and flecks of backdrop survive as speckle. Too high and pale parts of the product — white labels, chrome, light packaging — begin turning transparent. There is no universally correct number: it depends on how uniform the sweep was and how noisy the photo is. Let the tool detect the background first, then raise tolerance only until the backdrop disappears and stop.
Edge cleanup
At the boundary, a pixel is physically a mixture of product and backdrop. Removing the background does not remove that mixture, so a product shot on black and composited onto white keeps a grey rim. Edge cleanup un-mixes the pixel: given the estimated transparency and the old backdrop colour it reconstructs the underlying colour before compositing. Applied everywhere it would damage texture, so it only runs where a pixel is genuinely semi-transparent.
Edge softness
A one-pixel feather on the alpha channel. It hides the stair-stepping that a hard threshold produces on curved edges. Push it too far and the product starts looking like a sticker, with a soft translucent outline that survives onto the marketplace page.
Product size in frame
About 85 percent is the number most marketplace guidance settles on, and it reads well everywhere: it leaves an even margin on all four sides instead of letting the product run into two of them. At 100 percent a wide photo touches the left and right edges and all the leftover space piles up above and below, which looks cramped; a tall photo does the mirror image.
Maximum file size
Most upload forms cap somewhere between 200 KB and 5 MB, and marketplaces recompress what you send anyway. For a 2000 px product image, 200 to 400 KB is a sane target: comfortably sharp, fast to load, and well inside every limit worth caring about. Targets below about 50 KB for a photograph mean visibly throwing detail away.
What this pipeline will not do
- Cut out a busy background. Colour analysis cannot separate a product from a room. Use a segmentation model, and understand that the good open ones are licensed for non-commercial use only.
- Recover a soft or under-resolved original. Enlarging a 600 px photo to 2000 px invents no detail; it produces a blurry file that still fails quality checks.
- Fix an off-white source. If the original was shot under a warm bulb, the product itself carries the cast. Replacing the background with pure white makes that cast more obvious, not less.
- Rescue reflective or transparent products. There is no background to distinguish from the object when the object mirrors its surroundings.
Questions
Are my photos uploaded anywhere?
No. Every step reads and rewrites pixels through the browser's canvas API. The files are never transmitted, and the page continues to work if you disconnect from the network after it loads.
Why is the product a different size on different photos in the same batch?
Almost always because the trim step was switched off. Without it the frame proportion is applied to the whole original photo, including whatever empty margin each shot happened to have, so two photos of the same product come out at different apparent sizes. Leave trim the empty margin switched on unless you have a specific reason not to.
Can I use only one part of the pipeline?
Yes. Each of the three steps has its own switch. Turn off the background step to resize and compress existing photos that are already on white; turn off the frame step to get trimmed, white-background cutouts at their original size; turn off everything except step 3 and it behaves as a plain size-targeting compressor.
Why does the tool reduce dimensions when I only asked for a smaller file?
Because at some point quality alone stops helping. Once the encoder is at its sensible floor, the only lever left is pixel count, and file size scales roughly with pixel count. If you would rather have a failure than a smaller image, switch off shrink dimensions and raise the target instead.
How do I get a transparent PNG instead of white?
This page always composites onto a pad colour, because that is what marketplaces ask for. For a transparent cutout, use the white background tool on its own and choose transparent as the new background with PNG output.
Does it work on HEIC files from an iPhone?
Only if your browser can decode them. Chrome and Edge cannot read HEIC directly; Safari on macOS and iOS usually can. If nothing loads, convert to JPEG first.
Related
- Make a product photo background white — the cutout step on its own, with a transparent option
- Resize product photos in batch — the frame step on its own, for photos already on white
- Compress an image to an exact file size — the compressor on its own, with dimension limits
- Convert image format — JPEG, PNG and WebP batch conversion
- Guide: getting a pure white background for product photos
- All tools