Headshot background to neutral grey: cutout plus Pillow or an AI edit

To change a headshot background to neutral grey, the cheapest and most faithful route on Sume is a cutout followed by a local composite: POST /v1/rmbg-1.0/remove costs $0.0225 per image, returns a PNG with alpha, and a few lines of Pillow put that PNG on a grey canvas. The face is never regenerated, so it stays pixel-identical. An AI edit with ideogram/ideogram-v4.5 costs $0.075 at medium and can also relight, but it re-renders the image and can drift.

If the headshot is for a staff page, a badge or a press kit, identity matters more than polish. This post compares the two routes with costs for 100 headshots and gives the composite code.

Route A: cutout plus composite

Sume RMBG 1.0 takes a public HTTPS image_url and returns a mirrored PNG with alpha. The price is the same regardless of image size, $0.0225 per image after Sume's margin. Submit with mode: "async" and poll the job, or use sync for a bounded wait of up to 30 seconds; the job result endpoint is the standard one described in the Sume jobs docs.

Hair is the hard part of any cutout. Fine strands and flyaways produce a soft edge, and a hard edge looks like a sticker. When you composite on grey, check the shoulders and the hair at 200 percent, and feather the alpha by a pixel or two if you see a halo. Because the background is a flat colour you choose, you can pick the grey that matches your brand instead of describing it in a prompt.

from PIL import Image

cut = Image.open("headshot_cutout.png").convert("RGBA")
bg = Image.new("RGBA", cut.size, (128, 128, 128, 255))
bg.alpha_composite(cut)
bg.convert("RGB").save("headshot_grey.jpg", quality=95)
print("saved", cut.size)

Route B: an Ideogram 4.5 edit

ideogram/ideogram-v4.5 edits when you pass input_references: the first image is the one that is edited. An edit without aspect_ratio keeps the shape of the source. Ideogram says 4.5 is its most precise edit model and that it reduces drift across multi-turn edits. That is useful for a single background swap, but the model still produces a new image, so skin texture, collar edges and small details can change.

Use this route when you also want a lighting change, a softer backdrop with a gradient, or a studio look that a flat cutout cannot give. Say in the prompt that face, hair, clothing and pose stay unchanged, and then compare the face region with a pixel diff.

Cost for 100 headshots

The grey composite runs on your own machine, so only the cutout is billed. The edit rows use Sume's per-image prices for Ideogram 4.5 (provider list times 1.25).

Edge cases that decide it

Three things trip up headshot batches. Glasses with reflections: a cutout keeps the reflection of the old room in the lenses, so a grey background with a bright window reflected in the glasses looks odd; an edit can clean this, a cutout cannot. Hair close to the backdrop colour: a dark-haired person against a dark wall gives the cutout little contrast to work with, so look at the hairline first. Low-resolution sources: a 300 pixel avatar gets a ragged cutout edge, and enlarging first with the Sume Image Upscale 1.0 route costs $0.20 per image, nearly ten times the cutout itself.

So the order for a clean pipeline is: check the source size, cut out, composite, review at 200 percent, and only then decide whether any image needs the pricier edit route. In a batch of 100 where eight need the edit at medium, the mixed cost is 100 x $0.0225 + 8 x $0.075 = $2.85.

Which route to choose

Choose the cutout when the person must look exactly like the person: staff directories, ID-style photos, conference badges. Choose the edit when the source background is cluttered in ways that confuse a cutout, such as a hand-held plant crossing the shoulder, or when you want a re-lit studio feel.

One caveat on cutouts of images that already have transparency: Sume's RMBG flattens an input with large fully transparent areas onto mid-grey #808080 before sending it for removal, to keep the edges clean, and it still counts as one billed removal. If your source is a PNG with a transparent background already, skip the cutout and composite it directly.