Bedrock background removal (Nova, Titan) vs Sume transparent output

Cutting a product out of its background is a one-call job on Amazon Bedrock. Nova Canvas has a BACKGROUND_REMOVAL task that returns a PNG with 8-bit transparency, and Titan Image Generator v2 has a background removal feature that returns a transparent background (Amazon docs, read 2026-10-06). The Sume Image API has a different shape, so port with care.

What changes

On Bedrock the cutout is its own task: image in, same image out without a background. On Sume the transparent background is an option on a generation, so the model redraws the picture rather than masking pixels. For a packshot where the product must stay pixel-exact, that matters. Sume also has a separate background removal tool priced per image, covered in the post on cleaning 200 product photos.

Verify the alpha channel

Whichever service you use, check that the file really has transparency. This test fails on a flat white background.

import io, os, requests
from PIL import Image

r = requests.post("https://api.sume.com/v1/images", timeout=90,
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={"model": "openai/gpt-image-2.5", "background": "transparent",
          "output_format": "png", "prompt": "A single red apple, studio lighting."})
r.raise_for_status()
img = Image.open(io.BytesIO(requests.get(r.json()["data"][0]["url"]).content))
print(img.mode)
alpha = img.convert("RGBA").getchannel("A")
print("transparent pixels:", sum(1 for a in alpha.getdata() if a == 0))