1,000 transparent PNGs: gpt-image-2.5 background vs model plus RMBG

The cheapest transparent PNG on Sume is gpt-image-2.5 with background: "transparent" at low, $0.0074 each or $7.375 per 1,000. Every other image row has no background field, so you generate then run RMBG at $0.0225 per image, which adds $22.50 per 1,000.

In the Image API table background (auto, transparent, opaque) is a gpt-image-2.5 parameter; other models return 400 if you send it.

Cost per 1,000 transparent PNGs

RMBG rows add the removal price to the generation price.

The direct route

import os, requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "openai/gpt-image-2.5",
        "prompt": "Single enamel pin shaped like a mountain, centered, no shadow",
        "quality": "low",
        "background": "transparent",
        "output_format": "png",
        "image_size": {"width": 1024, "height": 1024},
    },
    timeout=60,
)
print(r.status_code)
body = r.json()
if r.status_code == 200:
    for item in body["data"]:
        print(item["url"])
else:
    print(body)

When RMBG is the better route

Use RMBG when the pixels come from a photo you already have, when you want a model that gpt-image-2.5 cannot replace, or when you want the generation and the cutout to be separate retryable steps.

Decision list

• Fully generated asset, any size: gpt-image-2.5 with background: "transparent".
• You need a particular model's look: generate with it, then RMBG for $0.0225.
• You have photos already: RMBG alone, no generation step.
• Read GET /v1/images/models/{model_id}/endpoints for the live pricing line before a large run: prices are catalog data and can change.

A worked example: 250 icons

A 250-icon set at 1024 on gpt-image-2.5 with background: "transparent" costs $1.84375 at low, $4.125 at medium and $16.46875 at high, using the per-image prices in the table. The table prices are for bare text prompts; a typical 140-character prompt adds roughly $0.00025 per image in prompt tokens, which is $0.0625 across 250.

Failed generations are not billed, so a batch that partly fails costs only the images that completed.