Logo screenshot to a transparent PNG: RMBG or GPT Image 2.5 background

To turn a logo screenshot on a white background into a transparent PNG, use Sume RMBG: POST /v1/rmbg-1.0/remove with a public HTTPS image_url, billed at $0.0225 per image regardless of size, and it returns a PNG with alpha. The other route is an edit on openai/gpt-image-2.5 with background: "transparent", which costs $0.009375 to $0.0835 per image in Sume's pricing fixtures but redraws the logo instead of cutting it out.

For a logo the difference matters. A cutout keeps the exact pixels you had. A redraw produces a new logo that looks like yours, which is not the same thing as your logo.

Why a redraw is risky for logos

Brand marks are precise shapes: a particular curve, a particular spacing between letters, a colour that must match a guideline. A generative edit treats the screenshot as a reference and paints a new image, so a wordmark can pick up an extra serif, a gradient can shift, and a thin outline can thicken. Each of those changes is small and each is a brand violation.

There is also a cost illusion. At low and medium quality the GPT Image 2.5 edit is cheaper than the cutout, and that makes it tempting. But you pay for a result that you must then compare with the original at pixel level, and if it differs you have paid for an unusable file.

RMBG for a logo

RMBG is designed for photographs, so a flat logo is an easy case, with one exception: white or near-white parts of the logo that touch the white background can be removed with it. A white letter inside a coloured badge is fine; a white outline on a white screenshot is not. If your logo has white parts, put it on a mid-tone background first or start from the source file if you have it.

Sume's RMBG flattens an input that already has large transparent areas onto mid-grey #808080 before removal, and still bills one removal. For a screenshot with a solid white background that does not apply, but it explains why a re-cutout of an existing transparent PNG can behave differently from a first cutout.

curl -X POST https://api.sume.com/v1/rmbg-1.0/remove \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://example.com/logo-screenshot.png",
    "mode": "async"
  }'

Cost for 50 logos

The RMBG price is flat per image. The GPT rows are the billed amounts for one 1024x1024 edit with one input image.

Check the alpha

Open the result on a dark background and on a light background and look at the edge. A halo of white or grey pixels around the logo means the cutout kept some of the old background. Zoom in on thin lines and small text. If the edge is rough, upscale before the cutout rather than after, since an upscaled logo gives the cutout more pixels to work with; the Sume upscale route is $0.20 per image, so use it only on the logos that need it.

For anything that goes into brand guidelines, prefer the vector source over any raster route. Use the cutout for the cases where you only have a screenshot, such as an old partner logo for a slide.

When to pick which

Pick RMBG when the logo already exists as pixels and must stay identical: partner logos for a slide, a screenshot for a press page, a badge from a scanned flyer. Pick the GPT Image 2.5 route only when you want a new interpretation, for example a rough sketch turned into a clean mark, and accept that you will review the result by eye.

Either way, keep the original screenshot next to the output and name the files so the transparent version is obvious. Tell whoever receives the file that it is a raster cutout, because a vector source remains the right deliverable for print.