Amazon doesn't ask for a "background that looks white": it asks for exact 255,255,255 and breathing room around the product. Here it's forced, framed and verified by re-reading the file we deliver.
"White by eye" doesn't exist for Amazon's validator: either the pixel equals 255,255,255 or it doesn't count.
You upload the real photo of your product and the server does the work with sharp and ZERO artificial intelligence. First it estimates the background by sampling the photo's perimeter. Then it separates it with a flood fill that enters from the whole edge, with adjustable color tolerance in three levels (Soft, Normal and Aggressive) so you can fine-tune when the product and background look alike. Then it cleans the outline halo, which is what gives away a bad crop.
With the product isolated, the background is forced to (255,255,255) as a VALUE, not a visual impression: all three channels are written to 255. The product is framed deterministically to fill 86% of a square 2000 × 2000 px canvas (you can lower it to 85% or raise it to 90%), with a safety margin so it never touches the edge. You're also warned if the product was already cut off by your original photo's edge, because that can't be fixed by cropping: it's fixed by reshooting.
And here's the part almost no one does: verification doesn't run on the in-memory intermediate result, it runs by RE-READING the binary you're going to download. The pixels of the perimeter band are counted and required to be pure 255. If the JPEG, even at maximum quality, dirties that perimeter with compression artifacts, the file IS REWRITTEN AS LOSSLESS PNG and re-measured. Pure white isn't promised: it's guaranteed because it's measured.
If your photo's background isn't separable, you don't get a mediocre image disguised as compliant. You're told why, shown how to reshoot the photo, and offered the option to audit the one you already have published. The whole process is visible in a before/after comparator with a slider, and the report comes with figures: percentage of pure 255 pixels in the perimeter, product occupancy, dimensions, file size and the four margins.
Use it free without an account. If you want to save your result, you can sign up afterwards.
Save / Sign up255,255,255 as a value, not an impression
all three channels are written to 255 across the whole background. Not an "almost" white, not #FEFEFE, not the white of your screen. It's the value the validator measures.
Verified by re-reading the delivered file
the measurement runs on the final binary, not on the in-memory intermediate. It counts the 255 pixels in the perimeter band of the file you download.
If JPEG dirties the white, it's rewritten as PNG
compression artifacts around the outline can break the perimeter even if the white was fine. When that happens, the file is regenerated as lossless PNG and re-measured.
Deterministic framing at 86% of 2000 × 2000
occupancy isn't eyeballed. The product's bounding box is calculated and scaled to fill 86% of the square canvas (adjustable between 85% and 90%), with a safety margin so it never touches the edge.
Three levels of crop tolerance
Soft, Normal and Aggressive. When the product is light-colored and so is the background, one button changes the flood-fill color threshold and you see the result instantly.
Zero AI in the cropping
background estimated by sampling the perimeter, segmentation by flood fill from the edge, and halo removal around the outline. All with sharp, repeatably. Your product stays your product: no pixel is invented.
Before/after comparator with a slider
drag it and see your original photo underneath and the compliant image on top. What's compared is exactly what gets downloaded.
It doesn't sell you a bad image as a good one
if the background isn't separable, it says so. It explains what failed, how to reshoot the photo, and offers to audit the one you already have published instead of handing you a dirty crop.
Upload your photo and get the main image with verified 255 background, 86% framing and the report with figures. Creating the account is free.
In the features that use AI, responses are generated by artificial intelligence and are not a substitute for professional advice.